From-machine-learning-to-deep-learning--Advances-of-the_2024_Current-Researc

Current Research in Biotechnology 7 (2024) 100164
Contents lists available at ScienceDirect
Current Research in Biotechnology

journal homepage: www.elsevier.com/locate/crbiot
From machine learning to deep learning: Advances of the recent

data-driven paradigm shift in medicine and healthcare
Chiranjib Chakraborty a, 1, *, Manojit Bhattacharya b, 1, Soumen Pal c, Sang-Soo Lee d, *
a
Department of Biotechnology, School of Life Science and Biotechnology, Adamas University, Kolkata, West Bengal 700126, India
b
Department of Zoology, Fakir Mohan University, Vyasa Vihar, Balasore 756020, Odisha, India
c
School of Mechanical Engineering, Vellore Institute of Technology, Vellore 632014, Tamil Nadu, India
d
Institute for Skeletal Aging & Orthopedic Surgery, Hallym University-Chuncheon Sacred Heart Hospital, Chuncheon, Gangwon-Do 24252, Republic of Korea
A R T I C L E I N F O A B S T R A C T
Keywords: The medicine and healthcare sector has been evolving and advancing very fast. The advancement has been
Deep learning initiated and shaped by the applications of data-driven, robust, and efficient machine learning (ML) to deep
Machine learning learning (DL) technologies. ML in the medical sector is developing quickly, causing rapid progress, reshaping
Artificial intelligence
medicine, and improving clinician and patient experiences. ML technologies evolved into data-hungry DL ap
Medicine and health care
proaches, which are more robust and efficient in dealing with medical data. This article reviews some critical
data-driven aspects of machine intelligence in the medical field. In this direction, the article illustrated the recent
progress of data-driven medical science using ML to DL in two categories: firstly, the recent development of data
science in medicine with the use of ML to DL and, secondly, the chabot technologies in healthcare and medicine,
particularly on ChatGPT. Here, we discuss the progress of ML, DL, and the transition requirements from ML to
DL. To discuss the advancement in data science, we illustrate prospective studies of medical image data, newly
evolved DL interpretation data from EMR or EHR, big data in personalized medicine, and dataset shifts in
artificial intelligence (AI). Simultaneously, the article illustrated recently developed DL-enabled ChatGPT tech
nology. Finally, we summarize the broad role of ML and DL in medicine and the significant challenges for
implementing recent ML to DL technologies in healthcare. The overview of the data-driven paradigm shift in
medicine using ML to DL technologies in the article will benefit researchers immensely.
1. Introduction the FDA (food and drug administration) in 2017 (FDA approved no.
K163253), which was developed by Arterys Inc., USA. Subsequently,
In the recent era, medicine field is changing very fast. The paradigm several AI-based products were approved from time to time (Benjamens
shift in medicine was noted in the area of cutting-edge diagnostics of et al., 2020; Ebrahimian et al., 2022). The approval of AI technologies
diseases, advanced treatment procedures, targeted drug therapy, and boosted AI research. Significant AI-related algorithms have been pub
surgery with the recently evolved artificial intelligence (AI) technology lished during the last decade, demonstrating AI in medical science. The
like deep learning (DL) technologies. The recent progress of machine information about AI-related algorithms boosted AI-related research in
learning (ML) plays a crucial role in the paradigm shift in medicine. The medical science. DL is a subfield of ML that mainly uses algorithms of
ML is a form of AI that uses algorithms to predict outcomes precisely. In neural networks to predict outcomes more accurately. In medicine, DL-
medicine, these algorithms use medical data to provide new outcomes. based algorithms are more efficient and robust which are mostly used in
Presently ML systems are frequently used in the medical field (Rajkomar medical images to solve diagnostics challenges (Kurokowa et al., 1987;
et al., 2019; Goecks et al., 2020; May, 2021; Haug and Drazen, 2023). Castiglioni et al., 2021). In 2016, Gulshan et al. developed a DL-based
The first AI-based cardiac MRI software, “Cardio AI,” was approved by algorithm for identifying DR (diabetic retinopathy) using the
* Corresponding authors at: Department of Biotechnology, School of Life Science and Biotechnology, Adamas University, Kolkata, West Bengal 700126, India
(Chiranjib Chakraborty) Institute for Skeletal Aging & Orthopedic Surgery, Hallym University-Chuncheon Sacred Heart Hospital, Chuncheon-si, 24252, Gangwon-do,
Republic of Korea (Sang-Soo Lee).
E-mail addresses: [email protected] (C. Chakraborty), [email protected] (S.-S. Lee).
1
Authors contributed equally.
https://doi.org/10.1016/j.crbiot.2023.100164
Received 18 August 2023; Received in revised form 19 November 2023; Accepted 21 November 2023
Available online 22 November 2023
2590-2628/© 2023 The Author(s). Published by Elsevier B.V. This is an open access article under the CC BY-NC-ND license (http://creativecommons.org/licenses/by-
nc-nd/4.0/).
C. Chakraborty et al. Current Research in Biotechnology 7 (2024) 100164
development dataset of 1,28,175 retinal images. The DL algorithm medicine, and continuous monitoring, which is crucial in advancing the
showed high specificity and sensitivity (Gulshan et al., 2016). In the healthcare industry.
following year, using DNN (deep neural networks), Esteva et al. devel Although ML-based algorithms have enriched the different medical
oped an algorithm for the classification of skin cancer using a dataset of fields, at the same time, along with the advancements in ML-related
1,29,450 clinical images (Esteva et al., 2017). Again, in 2018, Rajpurkar algorithms, there is a considerable advancement in data science and
et al. formulated and validated a DL algorithm named ‘CheXNeXt’ for big data analytics in the field of medical science. Researchers are trying
chest radiograph diagnosis where the researchers used 420 images to to evaluate and find the correlations between real-world medical data
classify the clinical abnormalities in chest radiographs (Rajpurkar et al., sets and big data. Researchers use ML occasionally in clinical, multi-
2018). Thus, in three consecutive years, these three DL algorithms have omics, and pharmaceutical R&D (research and development) data in
created examples in medical diagnostics. this direction (Wang and Preininger, 2019; Ahmed, et al., 2020; Aliza
Looking back, it is noted in 1950, Alan Turing was the first to explain dehsani et al., 2021). Big data in the medical sector can transform data
the concept of simulating intelligent behavior and critical thinking using into knowledge. Datasets has been were analyzed for personalized
computers. After six years, John McCarthy coined AI. The computer medicine (Millman et al., 1989; Dicuonzo et al., 2022).
scientist defines AI as the engineering and science of creating intelligent Recently, ML, and DL models have used to detection of different
machines (Chahal and Byrne, 2020; Kaul et al., 2020; Yasnitsky, 2020). diseases and assessed through the randomized clinical trials (RCTs).
Several AI developments occurred between the 1970s and 2000s. In Wang et al. created a colonoscopy computer-aided detection system
adition to that, several computational problems were solved before 2000 (CADe) by DL-based platform to detect the adenoma. The colonoscopy
by computer scientists. Although, had calculation speed limitations, all with the CADe system was registered in a clinical trial on chictr.org.cn
these inventions occurred in the computer laboratory. In about 2000, (Clinical Trial ID: ChiCTR1800017675) (Wang et al., 2020). Hollon
the vanishing gradient problem of DL was appeared. Although in the et al. developed a diagnosis method for detecting brain tumors using
1950s, the concept of DL was first developed, its medicinal application deep CNN (convolutional neural networks). The technology was tested
was started in 2000 along with several limitations (Yang and Bang, through the multicenter clinical trial with many patients (n = 278)
2019; Kaul et al., 2020). However, in 2017, the first US FDA-approved (Hollon et al., 2020). Moreover, the US FDA and other advanced coun
cloud-based DL was developed, Arterys, an application in healthcare. tries have approved several ML or DL-based algorithms and medical
A DL-based medical imaging platform was also developed as Cardio AI, devices for the clinic. Aisu et al. have listed the ML and DL-based
which can quickly analyze cardiac ejection fraction through magnetic medical devices approved in Japan. They found 11 ML and DL-based
resonance, maybe within a second. In 2007, the IBM developed an open- devices received regulatory approval. Among them, five were related
domain question-answering computation system entitled as ’Watson,’ to gastroenterology, and six were related to radiology (Aisu et al. (2022).
usually called ’IBM Watson.’ However, another technology was devel This article discusses the journey from ML to DL and data-driven
oped, named DeepQA, to give evidence-based medicine. Arterys, Car patient care. In the review article, we illustrated the ML and DL tech
dioAI, IBM Watson, and DeepQA were excellent examples of AI-based nologies in medicine in two directions: first, advancement in data sci
medical technology which were developed between the year 2000 to ence; second, chabot technologies in healthcare and medicine. To
2020 (Kaul et al., 2020). After 2020, the progress of AI-based technol illustrate the progress in data science, we discuss medical images data,
ogies in the medical field was noted very fast. ML and DL interpretation, data from EMR (electronic medical record) or
The contribution of AI and biosensors in the medical field is signif EHR (electronic health record), big data in personalized medicine, data
icant. In this connection, Qureshi et al. explored recent advancements in for medical research, and dataset shift in AI. Chabot technologies with
AI and biosensors within the medical and life sciences domains. The recently developed ChatGPT technology have been discussed. Finally,
paper highlights the crucial role of ML in enhancing medical imaging, we illustrated the broad role of ML and DL in medicine and future
thereby enabling more accurate diagnostic procedures. Furthermore, challenges.
the paper delves into the concept of precision medicine, which involves
tailoring treatments to individual patients. It emphasizes the contribu 2. The journey from ML to DL and data-driven patient care
tion of ML in analyzing large datasets to provide personalized care.
Integrating biosensors into the Internet of Things (IoT) ecosystem is also In 1943, a neural network model was developed by McCulloch and
discussed, highlighting their real-time monitoring capabilities for Pitts , which started the journey of ML (Wang et al., 2021). Early neural
physiological and chemical signals. Additionally, the paper addresses networks could have been more helpful and insightful. After its origin,
technological advancements, such as accelerated AI and edge ML models continued to evolve. Data-driven ML was built by the
computing. It outlines the challenges, potential issues, and prospects in research of Hinton. Subsequently, it improved as a DL technique and was
applying AI and biosensors in the healthcare industry, emphasizing their popularized in 2012.
transformative impact on medical science and patient care (Qureshi However, first, the different ML models have been used in the
et al., 2023). In a separate study, Phatak et al. proposed a comprehensive medical field to analyse the medical data (Table 1). At the same time, the
framework that combines wearable sensors, real-time location systems, evolved DL models were utilized to analyze the medical data in a more
and AI/ML algorithms for data collection and analysis in sports and accurate way (Table 2).
healthcare. A vital feature of this framework is the provision of real-time Recently, data-driven ML was popularized as a specific application of
feedback to users, which enhances data-driven decision-making pro AI. Using algorithms, data-driven ML permits computers to trained and
cesses (Pathak et al., 2021). Jin et al. delved into the applications of AI in improve from data and experience (Ng et al., 2020). Dinh et al. devel
diabetes management, specifically using closed-loop control algorithms, oped a data-driven ML approach for an automatic mechanism to identify
glucose predictions, and calibrations within continuous glucose moni patient’s risk of cardiovascular diseases and diabetes (Dinh et al., 2019).
toring systems. The study examines the challenges and opportunities Zhao et al. developed Chinese diabetes datasets (both type-1 and type
associated with individualized and proactive medicine (Jin et al., 2023). − 2 patients) for the development of data-driven ML models (Zhao et al.,
In another study, Lu et al. emphasized AI’s pivotal role in analyzing data 2023). Similarly, Liu et al. performed a data-driven population segre
collected from various biosensors. Their article underscores how AI gation analysis using a ML model for clinical use (Liu et al., 2023).
contributes to the long-term and in-situ monitoring of physiological Along with the popularization of ML techniques, data-driven DL is
information, aligning with the broader landscape of IoT technologies in currently gathering enormous attention for patient care. Oh et al.
healthcare (Lu et al., 2023a). Collectively, these articles reflect the developed a cluster-based deep reinforcement learning (DRL) approach
increasing significance of AI-driven biosensors in shaping the future of for the treatment of T2D (type-2 diabetes) using electronic health re
healthcare. They facilitate data-driven decision-making, personalized cords (EHR) of the South Korean population. The model shows high-
2
Table 1 Table 2
A list of different ML methods used for the recent data analysis of medicine and A list of different DL methods used for the recent data analysis of medicine and
healthcare. healthcare.
Sl. Significant concept of ML Description Sl. Significant concept of DL Description
No. No.
1. Bayesian learning This method combined the preceding 1. Generative adversarial The neural network with generative
knowledge information in addition to network approach,where two neural networks are
data, to accomplish ML. trained. One neural network, thegenerator
2. Ensemble learning Ensemble learning approaches build mainly provided with a set of randomly
many models and use the average of all generated inputs, and tasked with producing
the models to create predictions. samples. The second, the discriminatoris
Collective ensemble approaches include trained to generated, differentiate and real
the gradient-boosting, random forests, samples.
and with the stacking/meta-ensembles. Afterwards the two neural networks are
3. Unsupervised, supervised, and Unsupervised learning identifies the trained against each other, the resulting
semi-supervised learning structure, generally clusters, among the generator can be used to produce new
unlabeled data. Whereas, the supervised important examples
learning predicts labels or classes on 2. Neural network A ML approach mainly conceptualized by
future data, based on past data that holds biological neurons, the inputs are fed into one
labels/classes. Semi-supervised learning or more layers, and producing an specific
initially accomplishes unsupervised output layer
learning, and humans label structures that 3. Restricted Boltzmann The neural network having generative from of
found from the unsupervised learning. machine neural network that forms the building block
4. Dimensionality reduction It reduces the number of features or for support to many deep learning
attributes of a dataset by choosing vital approaches, having a single input layer and a
features or merging the features to single hidden layer. In this approaches no
capture variance in a dataset. Frequently connections developed between the nodes
used to increase the machine learning within each layer
models performance and to support 4. Recurrent neural One type of neural network with cycles in
visualization. network between the nodes inside a hidden layer
5. Deep learning Deep learning is the multi-layer artificial 5. Supervised learning The ML approaches with the goal of
neural networks that can learn the prediction of labels or consequences/outputs
complex non-linear functions. 6. Autoencoder The neural network where the training
It is too much useful for unstructured data objective is to reduce the error in between the
like speech, images, or text. But, usually it output layer and the input layer. This kind of
does not provide insights in to the aspects neural networks are unsupervised and are
of the data which are driving the frequently used for the dimensionality
functions. reduction
6. Federated learning This approaches used for incrementally
learning from data distributed in 7. Feed-forward neural The neural network that does not consist
numerous locations. It cannot be united network cycles between the nodes in the same layer
into a single dataset. It is useful when data 8. Denoising autoencoder The special type of auto encoder comprises a
are positioned in multiple clinical phase where noise is incorporated to the input
systems, or when the learning occurred during the stage of training process. The
from sensitive personal data. denoising step actions as smoothing and may
7. These are the supervised learning permit for effective use on input data,this is
Regression and classification methods. Regression predicts real-valued inherently noisy
outputs such as response to therapy. The 9. Variational One of the special type of auto encoder
classification can predicts discreet autoencoder generative AE learns a probabilistic latent
categories such as normal vs. diseased. variable model
10. Unsupervised learning ML approaches with the aim of pattern
identificationor the data summarization
quality performance when it resembles existing DRL models (Oh et al., 11. Convolutional neural The neural network consist with layers, where
2022). Therefore, a recent paradigm shift was constantly noted in data- network connectivity is preserves local structure. Even
the data meet the underlying assumptions
driven patient care using the ML and DL models, which are changing the performance is often good, and this networks
scenario of health care. can require less examples to trainefficiently.
Because they holds rarer parameters and also
offer improved efficiency
2.1. Progress in ML
12 Deep neural network The neural network consists with multiple
The speedy progress of ML was recorded in medicine and healthcare, hidden layers. Training occurs over the
and researchers have contributed to the progress immensely (Table 3). network, and accordingly such architectures
allow for feature construction to performed
The progression of ML algorithms and models has been incredible, with
combined with to optimization of the
the development of novel techniques to tackle different problems. An complete training objective
example is graph machine learning (GML), which represents a new 13 Deep belief networks The generative neural network with
classification of ML methods that harnesses the structural characteristics numerous hidden layers, which can be
of graphs and other irregular datasets to acquire effective feature rep acquired from combining of the multiple of
Restricted Boltzmann Machine
resentations of nodes, edges, or even entire graphs (Gaudelet et al., 14 Long short-term memory This special type of recurrent
2021). ML has applications in various sectors, including drug discovery (LSTM) neural network neuralnetworkhaving features that enable
and development, education, drought hazard monitoring and fore models to capture the longer-term
casting, and cancer research. In drug discovery and development, using dependencies
15 Multilayer perceptron One type of type of feed-forward neural
repurposed drugs through in vivo studies suggests that GML is on track
network having minimum one hidden layer,in
to become the preferred modeling framework within the field of
(continued on next page)
biomedical ML. GML has piqued the interest of the pharmaceutical and
biotechnology industries due to its capacity to model biomolecular
3
Table 2 (continued ) Table 3

Sl. Significant concept of DL Description
Recent performance of ML in medicine and healthcare.
No. Sl. ML model and its area of Remarks Reference
No. application
where each deeper layer is a nonlinear
function of every earlier layer 1. Weka-based classification It is used to diagnose breast (Bharati et al.,
16 Adversarial training The specialized process by which artificial algorithms on breast cancer disease using 2018)
training examples are nastily designed to fool cancer prediction various ML data mining
an neural network and then input as training tools and is considered a
examples to make the resulting fool an neural new aspect of medical
network robust progression. It was
17 Generative neural The neural networks that fall into this class of explored to evaluate the
network neural network can be used to produce data cancer disease dataset
similar to input data. These models can be collected from the UCI ML
sampled to produce hypothetical examples repository.
18 Data augmentation A procedure by which transformations 2. Logistic regression, and Data-driven diagnosis of (Bharati et al.,
doesn’t affect appropriate properties of the hybrid random forest and polycystic ovary syndrome 2020)
input data, and are applied to training logistic regression (RFLR) using the ML algorithms to
examples to upsurge the size of the training are applied to the a dataset freely available in
set diagnosis of polycystic the Kaggle repository.
ovary syndrome dataset
3. ML utilized in the field of The multiple ML (Chafai et al.,
structures, elucidate the functional relationships between them, and personalized medicine to algorithms utilized in 2023)
integrate multi-omic datasets. ML techniques have been implemented in integrate clinical data and clinical and genomic
genomic information of medicine for
research and education to augment teaching, learning, assessment, and
patients implementation of AI in
educational support. In one study, Cao et al. have highlighted the clinical medicine, drug
importance of geometric machine learning in several research applica discovery, and genomic
tion areas (Cao et al. (2022). Several other milestones and break medicine.
throughs in ML research are available. 4. ML driven approach for Focus on the ML-driven (Cammarota
large amounts of complex approach’s potential need, et al., 2020)
Sarker et al. explored the convergence between AI, the internet of healthcare information role, and limitations to
medical things (IoMT), and blockchain in healthcare. The research analyze complex
delves into the potential utilization of these technologies to enhance healthcare evidence of big
patient care through data-driven approaches. AI plays a crucial role in data sectors.
5. ML and its potential role ML refers to learning (Kilic, 2020)
this technology use as it facilitates the processing of extensive healthcare
in cardiovascular independently and making
datasets, leading to improved patient outcomes and the implementation healthcare accurate predictions in
of predictive analytics (Sarker et al., 2023). Bharati et al. conducted a cardiovascular healthcare.
study in breast cancer prediction utilizing various classification algo Additionally, applied in
rithms. Their work demonstrates the practical application of ML tech automated imaging
interpretation, natural
niques for early detection and diagnosis, which marks a significant language processing and
advancement from traditional ML to more sophisticated models in the data extraction from
healthcare domain. The WEKA tool’s comparative analysis offers valu electronic health records,
able insights into the performance enhancements (Bharati et al., 2018). and predictive analytics.
6. ML technologies in oral The ML is used for new (Leite et al.,
In a separate study, Bharati et al. focused their research on diagnosing
healthcare perspectives on 2020)
polycystic ovary syndrome (PCOS) through utilizing ML algorithms. diagnosing, classifying,
PCOS is a complex and challenging medical condition to diagnose and predicting oral
accurately, and this study showcases the potential application of ML in diseases, treatment
addressing such complexities within healthcare. The research reflects planning, and assessing
and predicting outcomes,
the shift from traditional ML to more advanced techniques in managing reducing the possibility of
intricate medical conditions (Bharati et al., 2020). Further, in another human errors.
study, Bharati et al. conducted a comprehensive review of the role of 7. ML used in pediatrics and ML can assist in healthcare (Matsushita FY
explainable artificial intelligence (XAI) within the healthcare domain. neonatology healthcare data, high-resolution et al., 2022)
medical imaging,
This multidisciplinary area of research is relevant in transitioning from
biosensors with continuous
traditional ML to advanced methods, as it highlights the importance of physiologic metrics output,
understanding the decision-making processes of advanced AI models and the OMICS science for
(Bharati et al., 2023). In the domain of data-driven care, Ibtisum S. pediatrics and
conducted a comparative study on various big data tools. This research neonatology.
8. Utilization of ML in ML used in laboratory (Obstfeld AE,
holds significant value in handling and processing the vast amounts of clinical hematology hematology will increase 2023)
healthcare data required for data-driven patient care (Ibtisum, 2020). In laboratory standardization and
another study, Rahmani et al. provided a comprehensive review of the efficiency by reducing staff
applications of ML within the field of medicine. This paper covers the involvement in
automatable activities and
evolution of ML within healthcare, ranging from its applications in di
fewer resources.
agnostics to the associated challenges and limitations. By understanding 9. ML application in The ML also contributes to (Uche-Anya
the applications and limitations of traditional ML within healthcare, this astroenterology and health imbalances in et al., 2022)
paper acts as a stepping stone towards adopting more advanced tech hepatology gastroenterology and
niques, thereby paving the way for a data-driven approach in patient hepatology, including the
diagnosis of oesophageal
care (Rahmani et al., 2021). Rubinger et al. have also discussed the role
cancer, management of
of ML and AI in research and healthcare. They emphasized the inflammatory bowel
increasing adoption of AI technologies within healthcare, showcasing disease, liver
their potential to revolution patient care and research. It serves as a (continued on next page)
4
Table 3 (continued ) Table 4

Sl. ML model and its area of Remarks Reference
Recent performance of DL in medicine and healthcare.
No. application Sl. DL model and its area of Remarks Reference
No. application
transplantation, and
screening of colorectal 1. The application of DL Such a model helps to (Chakraborty
cancer. strengthen ChatGPT in the recognize the structure of et al., 2023a)
10. ML strengthen in the ML algorithms are offered (Woodman RJ drugs and its target proteins, the structural
geriatric medicine to improve diagnosis, risk and Mangoni discovery, drug domain in the protein, the
stratification, and AA, 2023) development process drug-binding pocket, and
individualized approaches the active site of the
to patient management, protein-based drug target.
specifically in geriatric 2. DL for cancer diagnosis, DL is used in the analysis (Tran, 2021)
medicine. prediction and treatment of the complex biology of
11. ML supportive to cardiac ML technical and clinically (Ostberg, 2021) cancer, methylation and
surgery applied to cardiac surgery transcriptomic data, and
comprises diagnostics, histopathology-based
surgical skill assessment, genomic inference.
postoperative 3. DL strengthen ChatGPT It provides error-free, (Pal et al.,
prognostication, for biomedical more accurate, and 2023a)
augmenting intraoperative engineering and research updated information and
performance, and advances the biomedical
accelerating translational engineering design of a
research. medical device.
12. ML domain in It contributed to (Adamson AS, 4. ChatGPT support the It offers greater (Chakraborty
dermatology dermatologists in Smith A, 2018) application in medical applicability in medicine, et al., 2023b)
diagnosing and treating science architecture and training
skin diseases, thereby methods, medical
improving patient care. diagnosis and treatment,
and research ethical
issues.
testament to the ongoing evolution in this field, highlighting the shift 5. DL based support system The DL was performed to (Ting DSW
towards DL, big data analysis, and data-driven approaches in healthcare in ophthalmology detect diabetic et al., 2019)
retinopathy and
(Rubinger et al., 2023). Despite the substantial progress in ML research
prematurity retinopathy,
and education, numerous challenges persist and necessitate attention. glaucoma-like disc,
Ethical considerations, privacy concerns, and the imperative need for macular edema, and age-
effective collaboration between humans and AI are among the obstacles related macular
to integrating AI into education (Kumar et al., 2023a). degeneration. Moreover,
it uses optical coherence
tomography and visual
2.2. Progress in DLmodels in healthcare fields.
6. Deep learning (multi- Dl is accomplished to (Krittanawong
layered neural networks) automate medical image and Johnson,
A few significant contributions and developments of DL in medicine based application for interpretation, identify 2019)
and healthcare are summarized in this subsection. The noted re cardiovascular medicine novel phenotypes,
searchers’ contributions to the development and progress of DL in enhance clinical decision-
making, and select better
medicine and healthcare have been incredible (Table 4). As it is known,
treatment pathways in
DL models, which are also referred to as DNN, have an essential complex cardiovascular
advantage, i.e., these attempt to resemble human brain functioning and diseases.
can comprehend highly complicated patterns in vast datasets. DL can be 7 DL for COVID-19 Dl has been used for (Liu et al., 2022)
employed in several areas, such as health monitoring, image investiga diagnosis and prediction thoracic imaging to
diagnose, predict, and
tion, prediction of protein interactions, and detection of viruses. With manage COVID-19
DL, many unsupervised datasets can be comprehended and analyzed. patients with moderate to
Therefore, it undoubtedly becomes a precious tool for undertaking big severe symptoms.
data analytics where the raw datasets are mostly not categorized or 8 DL model (deep neural This model integrates (Pandey et al.,
network) for multidimensional 2021)
labeled. Najafabadi et al. studied the use of DL in handling notable issues
echocardiographic echocardiographic data to
with big data analytics, such as extracting complicated patterns from assessment of diastolic identify distinct patient
enormous datasets, tagging data, retrieving swift information, semantic dysfunction subgroups having heart
indexing, and doing more straightforward discriminatory jobs. This failure with preserved
group further explored the requirement of more research in the DL ejection fraction.
9. DL system for differential The deep learning system (Liu et al.,
domain to tackle the exact challenges that crop up from time to time in diagnosis of skin diseases (DLS) to provide a 2020a)
big data analytics (Najafabadi et al., 2015). In another study, Kaul et al., differential diagnosis of
in their book chapter, highlighted the contribution of DL techniques skin conditions using
when vast volumes of medical data need to be processed and analyzed. 16,114 de-identified cases
(photographs and clinical
They noted the DNN working mechanism concerning big data analytics
data) from a
and the capacity of DNN in detecting and treating human diseases (Kaul teledermatology practice.
et al., 2022). The contribution of DL in the healthcare domain was also 10. DL approach used for Large and diverse datasets (Hammad and
assessed by Shamshirband et al. They studied DL models used in accurate eczema and of skin images are used to Pławiak, 2023)
healthcare by considering novel network architectures and their appli psoriasis skin detection study the deep learning
approach to detect
cations. They highlighted the use of DL models in healthcare to unite DL multiple skin diseases and
technologies and healthcare interpretation (Shamshirband et al., 2021). improve the efficiency
Jin et al. made an elaborate study on DL use in COVID-19 research. Their and accuracy of
study concluded that DL has the potential not only to diagnose COVID- dermatological diagnosis.
5
19 but also to judge the progress and prognosis of the disease, suggest 2.3. The transition requirements from ML to DL
treatment plans, and assist federal governments in formulating intelli
gent steps to control and prevent the spread of the disease (Jin et al., The transition from ML to DL in the healthcare field is a multifaceted
2023). Another detailed study by Zvarikova et al. discussed the role of and transformative journey, encompassing many intricate challenges
ML and DL techniques, tasks involving computer vision, and IoT-based that require close attention. One of the primary challenges is the pro
health checking systems in handling COVID-19, i.e., detecting, testing, tection of data privacy and security. As the healthcare industry
treating, and preventive measures. It was pointed out that biometric increasingly relies on large and diverse datasets, it becomes crucial to
data could be deployed to diagnose remote COVID-19 (Zvarikova et al., establish strong encryption, access controls, and anonymization tech
2022). Concerning COVID-19 screening, Mondal et al. have also niques to safeguard sensitive patient information. Compliance with
mentioned the role of DL and federated learning in their review paper stringent regulations such as the health insurance portability and
(Mondal et al., 2023). In a different application area, Kumar et al. have accountability act (HIPAA) and the general data protection regulation
also highlighted the prospects of ML and DL in health monitoring sys (GDPR) is indispensable to mitigate data breaches and avoid legal
tems in their book chapter. They proposed a robust model that would consequences. Furthermore, achieving interoperability is a pressing
deliver superior confidence and precision compared to traditional ones concern. The seamless exchange of data between healthcare systems and
(Kumar et al., 2023b). In another interesting work, Wu et al. proposed an devices necessitates the utilization of standardized data formats and
IoT-enabled DL-based health monitoring unit that works in real-time. As protocols, such as the fast healthcare interoperability resources (FHIR),
a protocol, medical devices need to be put on, and the proposed unit to ensure smooth data sharing and integration. Without interoperability,
would gauge vital signs and utilize DL techniques to comprehend the potential advantages of DL may be hindered by fragmented and
precious information on health conditions. The case study was con isolated data, limiting the comprehensive understanding of a patient’s
ducted on Sanda athletes, and the monitoring system can deliver pre medical history. Ensuring the long-term maintenance and updates of DL
cision prediction during real-time transmission conditions, monitoring, models is another critical aspect. These models require ongoing atten
storage, and examination of datasets (Wu et al., 2023). DL has also tion to adapt to evolving medical knowledge and data patterns. Estab
played in information extraction of protein–protein interactions (PPI). lishing dedicated teams to maintain the models and address ethical
In this connection, Zhao et al. developed a DNN model that can extract implications and regulatory compliance is vital to prevent the obsoles
PPI information from biomedical literature. The model first utilizes the cence of DL applications. Gaining patients’ trust and acceptance depends
training method of auto-encoders for parameter initialization of a deep on transparency in the use of their data and obtaining informed consent.
multilayer neural network (DMNN) and second, the gradient descent Educating patients about the benefits and risks of DL in healthcare and
technique to train the DMNN model. The experimental results indicated how their data is utilized is crucial in fostering trust and acceptance of
better performance than multilayer neural networks (Zhao et al., 2016). these technologies. The issue of data ownership and governance is
In another work on PPI extraction from biomedical literature, Peng and multifaceted. Determining who owns healthcare data and has the au
Lu conceptualized a model that functions on a multichannel thority to access and utilize it is a fundamental question. Establishing
dependency-based convolution neural network (McDepCNN). The clear governance policies that address data ownership, access, and
McDepCNN model can utilize high-quality information procured from sharing can help resolve conflicts and ensure responsible data usage.
various channels, and it can easily generalize on various corpora, as well Promoting interdisciplinary collaboration is also essential for effectively
as record lengthy features of sentences (Peng and Lu, 2017). A detailed implementing DL in healthcare. It necessitates the seamless collabora
study on the status of PPI prediction with DL was undertaken by Sol tion of healthcare professionals, data scientists, and domain experts.
eymani et al., and they highlighted the latest DL technologies employed Encouraging interdisciplinary training and fostering a culture of
in predicting functions associated with proteins, PPI and their corre collaboration can bridge knowledge gaps and facilitate the integration of
sponding sites, binding among proteins and ligands, and design of pro DL technologies into healthcare workflows. Another significant chal
teins (Soleymani et al., 2022). Furthermore, DL technologies have lenge is determining liability in adverse events involving DL systems.
contributed to clinical image analysis considerably, and this field is When these systems are used in patient care, it is vital to establish legal
gaining increased attention with new research developments. As frameworks and guidelines that determine responsibility in case of
mentioned in the introduction section, many research groups have system failures, errors, or adverse outcomes. Regulatory compliance,
worked on employing DL methods in medical image analysis to tackle particularly with agencies such as the USFDA and the European Medi
diagnostic hurdles, retinopathy for diabetic patients, classification of cines Agency (EMA), is imperative to ensure that DL applications meet
cancer in the skin, and investigation of chest radiographs (Castiglioni rigorous standards for patient safety. Adhering to these guidelines is
et al., 2021; Gulshan et al., 2016; Esteva et al., 2017; Rajpurkar et al., vital to mitigate risks and legal liabilities associated with healthcare AI.
2018). Comprehensive reviews giving a detailed overview of the use of Ethical considerations play a significant role in the transition to DL in
DL techniques in comprehending and analyzing clinical photographs healthcare. These encompass issues related to bias in algorithms,
have been carried out recently (Suganyadevi et al., 2022; Shen et al., discrimination, and the equitable distribution of benefits. To address
2017; Razzak et al., 2018; Zhou et al., 2021). these ethical concerns, it is essential to establish robust ethical frame
DL has been proposed for used in solving global health problems like works, ethical oversight, and ongoing monitoring to ensure that AI
antibiotic resistance. Chakraborty et al. proposed to use DL to solve the systems are fair, transparent, and accountable. Finally, incorporating DL
problem of antibiotic resistance. They have shown a path to explain the in healthcare is contingent upon education and training. The assimila
prospect of DL to solve antibiotic resistance. Chakraborty et al. also tion of DL technologies necessitates healthcare professionals and data
proposed using DL in the different surgery processes (Chakraborty et al., scientists to possess the requisite knowledge and expertise. In this re
2022a). DL-enabled tools and methodologies have emerged as a gard, extensive training programs, workshops, and courses can bridge
powerful technique to characterize and learn from rapidly accumulating the knowledge disparity and ensure that the workforce is adequately
medical and healthcare data. Several researchers have used DL-enabled prepared for the forthcoming AI-centric era of healthcare.
tools in different fields of biological science research, especially in
medicine and healthcare (Chakraborty et al., 2022b; Chiu et al. (2020). 3. Advancement in data science in medicine in the journey of
Recently, DL-enabled ChatGPT has emerged as a powerful LLM means. It ML to DL: From clinical data to big data
has been used in different fields of medicine and healthcare, occasion
ally (Pal et al., 2023a). However, we have illustrated the application of Initially developed computers were quite slow and huge in the
ChatGPT in medicine and healthcare in a separate section in this article. 1950s. The storage capacity was meager, which was below 5 MB. The
first hard drive with 1 GB in capacity was introduced in 1980. The data
6
storage capacity was increased daily, and the computation speed was areas of medical imaging, such as understanding the image of
increased every day. Therefore, there is a huge improvement in the data mammography, lung cancer detection, detection in the area of
storage systems and computer devices such as memory, speed, and the ophthalmology, understanding in the area of gastroenterology, and
circuit. All these increase the data processing capacity. Even today’s cardiac function assessment.
pocket-sized devices are more powerful than the supercomputers of the
1980s. With the advancement of the computer system, both the cost per 3.1.1. Understanding the image and other data of breast cancer
unit of data storage and the device’s size was reduced considerably. In Understanding the images of breast cancer (BC) is an essential field
terms of data, today computers are having high processing capacity and of medical image data analysis using from ML to DL models. Scientists
faster more (Fig. 1a and Fig. 1b) (Leiserson et al., 2020; Haug and are especially using more efficient DL algorithms for medical and cancer
Drazen, 2023). image data. Wu et al. have used a deep convolutional neural network of
Clinical big data are transforming in the medicinal field. Medical DL for BC screening. Here, the researchers have trained and evaluated
data sets are analyzing from ML to DL, and the analyzed datasets were the image data using over 1,000,000 BC images. Here, they have per
interpreted from time to time (Table 5). Although, the ML algorithms formed the classification of BC images. The researchers provided a
need massive datasets to obtain an acceptable range, often requiring hybrid model and compared it with their previous one. Finally, they
millions of observations performance levels. Obermeyer and Emanuel have shown the hybrid model might predict the malignancy by a radi
stated that ML algorithms are high “data hungry” (Obermeyer and ologist more accurately (Wu et al., 2020). Using DL techniques, Shen
Emanuel, 2016). Therefore, a huge quantity of medical data can be used et al. have developed an improved detection method of breast cancer
to support the ML algorithms’ property of data hungriness. However, screening using mammography. In this study, researchers have used the
clinical data analysis like image data, big data, and dataset shift are the convolutional network method to classify mammograms. Here, they
crucial area in which ML, and DL analyze and predict. have used the approach of “end-to-end” training that leverages training
datasets efficiently and also the independent test set of FFDM (full-field
digital mammography) (Shen et al., 2019) In another study, for the DL
3.1. Medical images data and ML to DL interpretation model, Jaamour et al. used the divide and conquer-based technique for
mammography classification (Jaamour et al., 2023).
ML models have evolve and shown substantial improvements in Moreover, several researchers have used a hybrid model of DL
medical image data detection. The recent evolve DL has catalyzed a new techniques BC classification and detection. Altaf and Dewangan et al.
phase in the evolution of ML. For radiological image processing, DL have worked DL–based hybrid models for cancer detection (Altaf, 2021;
systems have shown significant improvements in accuracy in different
Fig. 1. The timeline and graphical representation show the advancement of computers. (a) the timeline depicts the development of computers from 1956 to 2023. (b)
the graphical representation shows computers’ advancement in size and storage capacity. these features of computer science cause the recent progress in medicine
and healthcare data science.
7
Table 5 Table 5 (continued )

Feature and variability of ML and DL for Medical data determining and out Sl. ML and DL Objectives Remarks Reference
comes forecasting. No. algorithm and
Sl. ML and DL Objectives Remarks Reference its area of
No. algorithm and application
its area of ML in systems owing to
application healthcare the
1. ML and the Examining the Development of (Jones et al. implementation
evolution of integration of ML support vector (2018)) of ML
human in human machine model 7. Finding out the Biomedical data ML and AI tools (Weber et al.,
healthcare healthcare for physiological missing link for combination and improvement to 2014)
data big biomedical analysis located analyze the
segmentation and data at heterogeneous biomedical data
analysis, sources for superior
prediction of clinical decision-
disease making.
progression and 8. ML for ML ML algorithms for (Rose, 2018)
diagnosis prediction in implementation addressing
2. Unplanned Efficient, safe, DL for (Cabitza electronic for better diverse clinical
consequences and effective, development et al., 2017) health record understanding of questions by
of ML humanistic care digital imaging, heterogeneous analyzing and
curating datasets, treatment effects finding the
integrative to apply precision nonlinear
heterogeneous medicine relationships in
data analysis, the electronic
identifying the health record
novel 9. Intelligent AI for progressive Operation of ML (Abidi and
associations, and health data health data and AI based Abidi, 2019)
remote analytics analytics analysis with the
monitoring and presence of
digital health data
consultations preprocessing,
3. AI, ML and data Implementation ML used for (He et al., choosing of
science, for of data science, finding patterns, 2019) algorithm based
laboratory AI, determining on expected
medicine and ML for the inefficiencies, outcome,
laboratory forecasting developing
medicine outcomes analytical
and taking models, and
accurate interpreting
decisions results
4. DL in health Reporting the ML for prediction (Naylor, 10. DL to transform Transform Operation of DL (Hinton,
care system unplanned of modeling in 2018) healthcare healthcare by ML for the digital 2018)
consequences due oncology, and application image analysis
to the application pattern 11 Genomics and Significant ML models to (Azencott,
of ML in present recognition in ML in improvements to address the main 2018)
healthcare pathology and precision address the challenges
systems radiology medicine genomic and of gene variations
5. ML knowledge Inspecting three ML methods for (Emmert- clinical and similarities
base with the main supports pattern Streib and data security among studied
ontology role integrating the recognition and Dehmer, problems patients
for pattern personalized development of 2018) 12. ML classifies Identification of ML for analyzing (Wong and
recognition in medicine into statistical models cancer novel tumor the histological Yip, 2018)
personalized routine clinical (sample size and classes data. The
medicine practices, which effect size). supervised ML
are categorized Knowledge base used for
phenotypically, of all present analyzing central
population size phenotype nervous system
and statistical categories and tumor type
analysis diseases. It genome-wide
organized the methylation data
clinical dataset of to detect the
experimental methylation
population size. patterns.
Specific software Unsupervised ML
platform for to examine the
statistical patterns in the
analysis of data sets for
higher- development of
dimensional categories
healthcare and classification
related multi- 13. Role of Primary diagnosis Proactive and (Sitapati,Kim
omics data electronic of chronic predictive et al., 2017)
6. Ethical Implementing Addressing the (Char et al., health record conditions by the involvement in
challenges of challenges of existing 2018) and integrated proper extraction healthcare
implementing ML in healthcare challenges in precision of clinical insights through AI, and
healthcare medicine in development of
8
Table 5 (continued ) Table 5 (continued )

Sl. ML and DL Objectives Remarks Reference Sl. ML and DL Objectives Remarks Reference
No. algorithm and No. algorithm and
its area of its area of
application application
personalized clinical decision conditions,

treatment support tool outcomes and
14. Fundamental Studied the ML for the data (Crown, risks
inference and implications of classification and 2019) 21. Precision Using ML to the It analyzed (Nayak et al.,
ML advancement in prediction in real medicine with electronic health patterns into the 2016)
observational world evidence to electronic records to subset of
research design maintenance of medical records produce population who
and healthcare clinical and personalized existing similar
databases regulatory medicine by clinical
decision-making converting it into phenotypes of
15. Use of Reporting Effecting of ML (Hersh et al., consistent risk complex disease
electronic cautions into the for over 2013) predictors, and
health record in existing whelming of integrating the
comparative healthcare present big data patient’s
effectiveness systems limitations in variability for
research healthcare prevention and
systems treatment of
16. Visualizing and Finding of future New ML, DL (Saheb and disease
analyzing the strands research, algorithms and Saheb, 2019) 22 Big data Application of big ML used for the (Raghupathi
knowledge comprising new advanced the big analytics in data data mining and and
structures of algorithms, data analytics for healthcare analytics in analysis Raghupathi,
health tracking tools and better- healthcare 2014)
informatics Internet of personalized 23 AI, ML and data Prognostic ML for healthcare (Gruson
Things-based treatment science, for modeling for data analysis and et al., 2019)
decision support laboratory improved optimization, cost
systems medicine collaboration reducing,
17. ML algorithms AI tools Human computer (Ngiam and among hospitals improving
and big data for improvement interaction -based Khor, 2019) without sharing efficiency of
healthcare based on AI and ML the patient’s data resources and
delivery incremental applications for and complying staff
learning to different clinical privacy
enhance the developments in regulation
predictive oncological 24 Healthcare Altering the ML implemented (Jacob et al.,
accuracies problems medical to precision 2023)
18. Intelligent Improving the DL for analyzing (Acs and solving with the treatment with medicine, which
digital diagnostic the images of Rimm, 2018) precision deference to the comprises data
pathology accuracy and whole-slide medicine individualized analysis and
efficiency with pathology features of storage for
the usage of ML patients determining the
19. ML in Investigating the The ML (Jiang et al., association
healthcare AI applications in algorithms used 2017) between disease
healthcare, and to extract and outcome and risk
their potential cluster data, and for identification
outcome in future perform the of patient
days principal characteristics
component and optimum
analysis, support treatment
vector machine to 25 Intelligent AI for advanced Application of ML (Abidi and
determine the health data health data and AI based Abidi, 2019)
model analytics analytics analysis with the
parameters, and addition of health
identify imaging data
biomarkers, preprocessing,
natural language choosing
processing for algorithm based
text classification on expected
and processing, outcome,
and DL for electro developing
diagnosis and analytical
diagnostic models, and
imaging interpreting
development results
20. ML in cancer ML applied to ML used to model (Kourou 26 ML in medicine Investigative the DL applied on the (Rajkomar
prognosis and detect crucial the progression et al., 2015) essential current electronic et al., 2019)
prediction features by and treatment of structural health record
predictive cancerous changes in the data to produce
modeling of conditions healthcare system associations and
complex and through that are essential meaningful data
heterogeneous examining to release the full for personalized
datasets for complex datasets potential of ML in treatment and
advancement and and revealing medicine disease diagnosis
treatment of their 27 Data analytics Analyzing the ML algorithm for (Stein et al.,
cancerous relevance and ML for electronic health the structured 2019)
9
Table 5 (continued ) 3.1.5. Assessment of the image and other data of gastroenterology
Sl. ML and DL Objectives Remarks Reference DL has also made advancements in gastroenterology, especially the
No. algorithm and improvement of colon and endoscopies. ML and DL models have assisted
its area of endoscopists and colonoscopists, making endoscopy and colonoscopy a
application more consistent tool for diagnosis. Scientists are trying to improve
disease record for and unstructured endoscopy techniques through these models.
identification in identification of big data analysis Yuan and Meng developed WCE (Wireless capsule endoscopy) to
electronic extensive range of leading to the
analyze the endoscopy images to recognize the polyps accurately. In this
health record medical identification of
conditions and wide range of study, the computer-aided diagnosis used DL-based techniques to detect
diagnosis medical and characterize the polyp images in a WCE video (Yuan and Meng,
conditions and 2017).
diagnosis Similarly, several studies of ML or DL models were noted for image
detection in colonoscopy. Wang et al. created a computer-aided DL-
Dewangan et al., 2022). However, other than the DL models, recently, based detection system. The study has tried to detect the ADR (adenoma
ML models have been used for breast cancer detection (Izci et al., 2023). detection rate) during the colonoscopy. The researchers have developed
the characteristics of adenomas and polyps identification. The study has
3.1.2. Assessment of the image and other data of lung cancer been registered for a clinical trial (Wang et al., 2020). Another study by
Lung cancer detection is important, and from ML to DL techniques Gong et al. developed an improved colonoscopy technique, the
were widely used to detect lung cancer. Recently, scientists have been “ENDOANGEL system,” performed through deep neural networks. To
using three-dimensional (3D) DL techniques for cancer and another understand the effect of the ENDOANGEL system, about 704 patients
disease diagnosis where a model has been reconstructed for the 3D view were randomly done the colonoscopy, where 349 patients used the
from an object using 2D image collection representing the scene. Ardila control colonoscopy (n = 349) and 355 patients (n = 355) used the
et al. use 3D deep learning models for end-to-end lung cancer screening. ENDOANGEL system. During the colonoscopy, the ENDOANGEL system
Their model simultaneously attains a high-level interpretation of 6,716 enhanced the significant adenoma yield (Gong et al., 2020). Similarly,
National Lung Cancer Screening Trial (NLST) cases and an autonomous Urban et al. developed DL-based models for real-time polyps detection
clinical verification set of 1,139 subjects (Ardila et al., 2019). Similarly, and localization through colonoscopy screening, and they observed that
Shimazaki and his colleagues have tried to detect lung cancer using the model’s accuracy was 96 %. In this study, the technique used ADR to
chest radiographs. In this direction, a DL-based algorithm was developed improve polyp detection. The DL model was created using convolution
using segmentation methods. The study has detected lung cancers with neural networks (CNNs) (Urban et al., 2018).
low mFPI (mean false positive indications) per image (Shimazaki et al.,
2022). 3.2. From ML to DL and pathology
3.1.3. Screening the image and other data of cardiac function During the journey from ML and DL, several models have been
Scientists are screening the image and other data of cardiac function developed and have made the most significant progress in pathology to
by applying ML and DL. These models are making essential advance understand new disease insights. In this direction, several studies
ments in the field. Ghorbani et al. tried to assess the echocardiogram developed models to detect the area of interest within slides and create
data to understand the cardiac function using DL. The researchers have the proper workflow. Similarly, studies have been performed to under
identified the local cardiac structures and evaluated the cardiac function stand disease inside using whole-slide imaging. Kather et al. tried to
and cardiovascular risk. They have developed a DL model, EchoNet, assess the microsatellite instability from histology images to understand
which can identify several cardiac parameters such as the ejection gastrointestinal cancer. The study uses the DL models for MSI (micro
fraction, systolic and diastolic volumes, etc. (Ghorbani et al., 2020). satellite unstable) screening, which may be very helpful for cancer
Fletcher et al. tried to assess the diastolic function from Echocardiog immunotherapy. Using whole slide images, Campanella et al. detect the
raphy using the ML technique, which showed evidence that ML algo area of interest using a learning-based DL system. From 15,187 patients,
rithms could correctly distinguish cardiac structures (Fletcher et al., the study used whole slide images dataset of 44,732 numbers to train
2021). Hu et al. recently developed a wearable ultrasound imager to and classify models (Campanella et al., 2019). Images of lung tissue were
detect cardiac function and understand cardiovascular health. The re tried to detect occasionally using histopathological images. Using DL,
searchers developed a DL model that mechanically understands the left Rajpu and Subasi et al. have developed lung cancer assessment tech
ventricular volume from the continuous image recording. It helps direct niques from histopathological images of lung tissue (Rajput and Subasi,
and real-time cardiac function estimation (Hu et al., 2023). 2023).
Besides ML or DL, single-cell pathology is essential for detecting
3.1.4. Assessment of the image and other data of ophthalmology cancer cells. Jackson et al. have tried to identify the intratumor
During the journey ML to DL, several significant models have been phenotypic heterogeneity using a single-cell pathology landscape. The
used extensively in ophthalmology. In this direction, several new algo study has quantified 35 biomarkers using several high-dimensional pa
rithms have been utilized from time to time. Scientists have primarily thology images (n = 720) from using tumor tissue from numerous pa
implemented deep learning systems (DLS) to classify and validate tients (n = 352) (Jackson et al., 2020).
different ophthalmological parts. Milea et al. used DLS to classify optical
disks, which were assessed into three categories: normal conditions, 3.3. Data from EMR or EHR
papilledema, or further abnormalities. The study uses excellent training
and data sets which use 14,341 data of photographs from 6779 patients. The systematic collections of EHRs can provide patient health in
The dataset contains about 9156 standard disk photographic datasets, formation. It contains some structured information and unstructured
2148 papilledema disk photographic datasets, and 3037 disk data set information. The structured information includes laboratory tests, di
with other abnormalities (Milea et al., 2020). Liu et al. developed a DL agnoses, procedures, and medications. Unstructured information in
system to identify the GON (glaucomatous optic neuropathy) with high cludes free text such as clinical physician notes (Wang and Preininger,
sensitivity and specificity. The DLS assessed the validation datasets (Liu 2019). Recently, DL models have been implemented to analyze the HER
et al., 2019). data (Table 6). To analyze using DL models, several architecture and
algorithms have been used, such as CNN (convolutional neural
10
Table 6
Different EHR models were developed using ML and DL to analyze the EHR data.
Sl. EHR ML and DL Patients Remarks Reference
No. model method numbers
1. Deepr Convolutional 300,000 It use to predict (Nguyen

neural the unplanned et al.,
networks readmission of 2017)
(CNN) patients
2. Doctor AI Recurrent 263,706 Predict the (Choi
neural medical codes et al.,
networks in future visits, 2016a)
(RNN) and duration
until next visit
3. RetainVis 63,030 Prediction of (LeCun
heart failure, et al.,
and cataract 2015)
4. eNRBM Autoencoders 7578 Outcome for (Tran
(AE) suicide rate et al.,
prediction. 2015)
Elixhauser
comorbidities,
diagnosis
linked groups,
emergency
attendances
and
admissions,
demographic
variables (age
10 years
intervals,
gender) are
applied
5. Med2Vec Feed-forward – Prediction of (Choi Fig. 2. Different architecture and algorithms used in ML and DL for medicine
neural medical codes et al.,
and healthcare.
networks in previous/ 2016b)
(FFNN) future visits
6. Ensemble Recurrent 216,221 It studied and (Rajkomar data points and clinical notes (Rajkomar et al., 2018). Huang and his
model neural calculate the et al., colleagues have used the ML approach to evaluate hospital stay timing
networks inpatient 2018)
and mortality using EMR data. Here, the researchers have used a CBFL
(RNN)/Feed- mortality,
forward neural unplanned (community-based federated machine learning) algorithm to evaluate
networks readmission for the EMRs (Huang et al., 2019). Using infectious disease EMR, Wang
(FFNN) 30-dy, long et al. have tried to develop a decision-making system in clinical infec
length of stay, tious disease using vast amounts of data. The data contains several cases
diagnoses
7. Deep Stacked 704,587 Used for future (Miotto
(n = 20,620) from seven types of infectious diseases analyzed through a
Patient denoising disease et al., DL model. In this study, they developed MIDDM (multiple infectious
autoencoder prediction. 2016) disease diagnostic model) with two methods (attention mechanism and
(SDA) Demographic residual network) to advancement of the system performance. The DL
variables (age,
model can predict and multi-class diagnosis of diverse infectious dis
gender, and
race), eases having improved accurateness (Wang et al., 2022).
diagnoses,
medications,
procedures, 3.4. Big data in personalized medicine
along with the
lab tests, free-
Big data, complex or too large dataset, primarily include three types
text clinical
notes also of data: structured, unstructured, and semi-structured. In biological
considerate science, several scientists have tried to transform big data for healthcare
8. DeepCare 7191 Applied for (Pham and personalized medicine. Marshall et al. have integrated DSM (Dy
Recurrent disease et al., namic Simulation Modelling) and big data toward personalized medi
neural progression, 2017)
networks medication and
cine. They have finally explained the synergies between the DSM and big
(RNN) unplanned data (Marshall et al., 2016). The big data management of genomic data
readmission is a considerable challenge. At the same time, it is very much essential to
prediction perform big data Analytics in the field of biomedical omics-data (gly
9. RETAIN 32,787 Prediction of (Choi
comics, metabolomics, transcriptomics, proteomics, genomics, etc.),
human heart et al.,
failure 2016c) which might help to improve personalized medicine (Hassan et al.,
2022). However, scientists are trying to analyze the big data for the
development of personalized medicine, and several review articles have
networks), RNN (recurrent neural networks), and FFNN (feed-forward been published to jot down the works and ideas in this field (Ristevski
neural networks) from time to time (Fig. 2) (Ayala Solares et al., 2020). and Chen, 2018; Cirillo and Valencia, 2019; Hulsen et al., 2019). Work
Rajkomar et al. used the DL models to predict multiple medical events has been performed on personalized medicine using big data. In this
from a considerable volume of HER data to improve healthcare quality work, Kang et al. developed a DCP model to forecast dental caries using
and drive personalized medicine. The data contains 46,864,534,945 a labeled dataset of 22,287 samples containing the features such as
11
gender, region, age, etc., from South Korea. In this study, researchers gRNA sequences. CRISPR/Cas9-based screening was used to organize
have used the Random Forest (RF) model of ANN (Kang et al., 2022). gRNA in the library (Liu et al., 2020b). Other than the CRISPR/Cas9, and
Recently, Zhang et al. have developed a hybrid algorithm based on ML in DL models have been utilized in DDR1 kinase inhibitors (Zhavoronkov
precision medicine for clinical decision support. Here, they have used et al., 2019), pathogenic infections (Zillmer, 1986), etc.
two methods: an improved BM25 algorithm and luster-based abstract
extraction. The study used two data sets namely TREC2018 and 3.6. Dataset shift in AI
TREC2017 to train the proposed model. Subsequently, to validate, the
one dataset (TREC2019) (Zhang et al., 2023). Scientists are trying to use Dataset shift is a challenging situation AI model where the joint
QSAR approaches and also developing new models during drug dis distribution of the training and test datasets changes. This situation is
covery using different ML to DL algorithms. Beam and Kohane illustrate known as “dataset shift” (Fig. 3). It is a significant problem in the AI
how ML algorithms can handle a large number of patient data. They also medical field (Finlayson et al., 2021). Several scientists indicated the
discussed DL algorithms, high rank on the ML algorithms, and help in medical dataset shift is noted due to changes in the medical dataset
diagnosis procedures using huge raw pixel data (Beam and Kohane distribution due to the change in the training dataset and test dataset.
(2018)). Recently, DL applications have been handling the challenges in Therefore, dataset shift will be challenging during the implementation
big data. In a review article, Najafabadi et al.. Shows the direction to of ML to DL models. Recently, Subbaswamy and Saria indicated that
application of using fast information retrieval, data tagging, semantic shifts are a prevalent condition when a model moves from the training
indexing, simplifying discriminative tasks, etc. (Najafabadi et al., 2015). phase to the test or deployment phase due to the change in different
conditions such as patient demographics, equipment, timing, disease
3.5. Data for medical research prevalence, treatment patterns, measurement pattern, and more (Sub
baswamy and Saria, 2020). Sometimes, it is referred to as changes of
Other than image classification, ML and its highly efficient DL-based classification, the concept of drift, the concept of shift, etc.
models are applied in protein structure and function prediction, anti
biotics discovery use and resistance, drug discovery and development 4. Chatbot technologies in healthcare and medicine
and several other areas of medical science.
Chatbot technologies, a computer program that can recognize
3.5.1. Protein structure and function prediction questions and provide automated responses, use AI and NLP (natural
Recently, DL models have been executed in the different areas of language processing) (Haug and Drazen, 2023; Jackson-Triche et al.,
protein structure and function prediction-related research, such as de 2023). The first chatbot was developed by J. Weizenbaum, a computer
novo protein modeling using DL (Greener et al., 2019), biochemical scientist, in 1966 at MIT, and it was named ELIZA. Since 1966, several
properties of protein using self-supervised autoencoders of DL models chatbot technologies have been developed in till today from time to time
entitled DiffNets (Ward et al., 2021), improved protein structure pre (Fig. 4a and Fig. 4b). Chatbot technologies have also been applied in
diction applying DL model like AlphaFold, (Senior et al., 2020), rational medical research (Table 7.).
protein engineering through deep representation learning algorithm Chatbots are conversational agents powered by AI that can guide
(Alley et al., 2019), understanding the interaction landscape from pro users to perform multiple tasks by responding in natural language (Jia,
tein molecular surfaces using geometric DL (Gainza et al., 2020). 2003). In several industries, including healthcare, they are becoming
more prevalent recently (Jovanović et al., 2020; Caldarini et al. (2022).
3.5.2. Drug discovery and development Chatbots can be used for self-diagnosis by patients who want to look
Drug design is an important field of medical research. Machine in through their symptoms before seeing a doctor. Online symptom
telligence and DL models have been implemented in the different arenas checkers fall far short of AI-powered chatbots like OpenAI’s ChatGPT,
of drug discovery and development, such as optimization of hit selection Microsoft’s Bing, and Google’s Med-PaLM regarding accuracy. Patients
in virtual screening driven by machine intelligence (Kumar and can utilize the chatbot to describe their symptoms, and the chatbot can
Acharya, 2022), lead optimization model development using DL (Green then provide a grim diagnosis and advise them to see a doctor. Health
and Durrant, 2021), lead optimization through the 3D ligand informa coaching is another chatbots application that offers patients personal
tion using AI/ML methods (Bleicher et al., 2022), the web server for ized advice and support (Mitchell et al., 2021). For instance, Ariana is a
drug discovery using DL-based web screening entitled DeepScreening chatbot that assists patients with therapy, provides information about
(Liu et al., 2019), the drug toxicity screening using DL models (Jimenez- their disease and treatment, and renders helpful guidance on managing
Carretero et al., 2018; Lee and Chen, 2021), etc. daily activities.
CoachAI, a conversational agent-assisted progressive health coach
3.5.3. DL models from antibiotics discovery to resistance ing platform, enables health intervention delivery to individual patients
DL-based models are used in diversified areas of research to study the and groups (Fadhil et al., 2019). Chatbots of this type can help patients
antimicrobial substance such as antibiotic discovery (Stokes et al. 2020); manage their health and wellness by recommending what to eat, how to
understanding the antibiotics resistance genes (Arango-Argoty et al., exercise, and how to reduce stress. Chatbots can also be used to improve
2018; Li et al., 2021); diagnosis and treatment of antibiotic resistance the readability in online mode for informed consent forms. In a recent
(Chakraborty et al. 2022); rapid identification of the resistance UTI experiment, participants were given information from consent forms
pathogenic bacteria (Fu et al., 2021), etc. using Rumi, an AI-powered chatbot (Xiao et al., 2023). Compared to
form-based engagement, the chatbot decreased the power imbalance
3.5.4. Several other areas of medical science among the participant and the researcher and improved consent form
ML and /DL-based models are helping in different another area of reading.
medical research, such as surgical practice (Chakraborty et al. 2022). Furthermore, chatbots can be used to provide patients with mental
Recently, CRISPR/Cas9-research has been an important area in healthcare. One example is Wysa, an AI-powered bot that helps people
biomedicine. Here, ML to DL moles have been developed from time to manage their emotions by utilizing cognitive behavioral therapy and
time. ML approach was used to understand the anti-CRISPR protein dialectical behavior therapy (Nicol et al., 2022; Devaram, 2020). The
families (Gussow et al., 2020). DL models were applied for two high- empathic chatbot is another type that can recognize the user’s emotional
adhere Cas9 variants to optimize the CRISPR guide RNA design (Wang state and alter conversations to help people receiving mental healthcare
et al., 2019). Similarly, ML algorithms were developed SeqCor to (Devaram, 2020). Physicians can interact with existing and potential
manage the resulting biases potential initiated by the arrangement of patients by adopting chatbots for remote patient monitoring (Sabour
12
Fig. 3. The figure shows the dataset shift. It is a challenging situation for the AI model where the joint distribution of the training and test datasets changes.
et al., 2023). Telehealth services can be delivered through facilities like optimization method. The ChatGPT can learn from a considerable
the SmartBot360 platform, and video chat can monitor patients from a amount of data using DL approaches, particularly transformer-based
distance (Dias et al. (2022). Health-designed chatbots allow patients to topologies, and produce human-like responses (Pournaras, 2023).
schedule visits with doctors and receive prescription reminders. These ChatGPT’s competence has an opportunity to alter statistical process
fantastic, constantly accessible health monitoring tools have impacted control (SPC) practices, instruction, and research. However, since the
healthcare (Fadhil and Schiavo, 2019). Therefore, chatbots can trans technology is still in the primary stages of development, it is prone to
form healthcare by providing patients with customized and efficient misuse or misinterpretation. The investigation has looked into
services. Numerous medical services are available, including self- ChatGPT’s capability to deliver code, describe critical concepts, and
diagnosis, health coaching, informed consent, support for mental generate knowledge pertinent to SPC practices, education, and research
health, and remote patient monitoring. Chatbots arerojectted to perform (Megahed et al. 2023). ChatGPT has become quite popular in the
a more prominent function in the healthcare industry as they develop. medical field due to its capacity to simplify radiological reports, facili
tate communication between patients and healthcare providers, and
5. ChatGPT: A recent chatbot technology developed using DL, generate academic papers at an acceptable standard (Sedaghat, 2023).
LLM and NLP models Medical students can practice their communication skills and diagnosis
methods in a secure setting using ChatGPT, replicating patient en
An intense effort was put forth for several years by research and counters (Amri and Hisan, 2023).
development teams of DL- based chatbot and chatbots of specific capa Additionally, ChatGPT can help researchers and medical students
bilities were introduced. In 2022, a unique Chatbot named ChatGPT was read and write academic articles by effectively summarizing a given
launched by OpenAI to generate informal talk remarkably similar to topic and producing an unbiased abstract. For patients and healthcare
humans. The initial ChatGPT model was based on GPT-3.5, and now the teams to better comprehend radiological findings, ChatGPT can be uti
updated ChatGPT version runs on GPT-4. ChatGPT is an advanced lized to translate the records into plain language (Lyu et al., 2023).
language-based bot that has changed contemporary natural language ChatGPT (with GPT-4) may even produce fictitious medical images,
processing and human–computer interaction (Biswas, 2023). ChatGPT such as electrocardiogram graphs and X-rays, free of consent and
uses deep neural network to produce human-like text (Hashana et al., copyright, allowing medical students to practice and improve their
2023). Large language models (LLMs) use recurrent neural network interpretation abilities. By providing medical advice and medical in
(RNN) models to comprehend text, a DL model. Using transformer ar formation databases, including symptom checks, fitness, and health
chitectures, it can use trillions of parameters. The transformer topology counseling, ChatGPT serves an essential purpose. The system also pro
was developed in 2017, a breakthrough in this direction. The trans vides a few unique medical functions, such as appointment scheduling,
former is a deep bidirectional transformer entitled Devlin and her col notification alert, and reminders for regular medication to satisfy the
leagues (Devlin et al., 2018; Benjamens et al., 2020; Ebrahimian et al., demands of patients or users (Chung and Fong (2014)). ChatGPT can
2022). It works with GPT model methodology, and this GPT is an enhance mental health by simulating patient interactions and assisting
extensive language model that has been trained on tons of data and can with mental health issues (Hisan and Amri, 2023). In addition, social
be used for many multiple tasks, which includes text creation, question networking services like ChatGPT can support communication and
answering, and language translation (Joublin et al., 2023). Reinforce furnish resources for college students’ mental health (Vornholt and De
ment learning from the human feedback concept works in ChatGPT, and Choudhury, 2021). While discussing ChatGPT’s benefits, it is essential to
the DL system’s output is adjusted utilizing the proximal policy mention that ChatGPT’s use in clinical settings remains challenging, and
13
Fig. 4. The timeline and graphical representation show different chatbot technologies and their origin time. (A) The timeline shows different chatbot technologies
and their origin time point. (B) It depicts different chatbot technologies.
its performance in the medical field is comparable to that of third-year they suggest after years of experience in their respective medical spe
medical students who achieve marks that are just below the passing cialties. There are also worries regarding the security and privacy of
standard (Gilson et al., 2023). More developments are needed to achieve patient data while using ChatGPT. Before implementing ChatGPT in
its widespread medical use (Li et al., 2023). Despite ChatGPT’s possible clinical practice, it is critical to investigate and resolve all these ethical
benefits for the healthcare system, its application has genuine ethical issues to a certain acceptable standard.
concerns. Utilizing ChatGPT, the potential for misusage of academic
articles is one of the crucial problems in this direction (Hisan and Amri, 6. The broad role of ML and DL application in clinical practice
2023). and healthcare management
It is very clear that ChatGPT has been applied in different fields of
medicine and healthcare, from medical education to research (Chakra In the medical domain, ML and DL is playing crucial roles across the
borty et al. 2023). It has been noted that the tool has been applied to medical field, which includes diagnostics, therapeutics, patients’ health
diverse branches of medical sciences, such as oncology, radiology, management, administration and regulation of hospitals, and countries’
ophthalmology, orthopedics, rheumatology, etc. (Chinnadurai et al., regulatory activities of healthcare management (Fig. 5 and Fig. 6) (He
2023; Ramamurthi et al., 2023: Bajaj et al., 2023; Chakraborty et al., et al., 2019; Jassar et al., 2022; Rajpurkar et al., 2022; Haug and Drazen,
2023a). It is also used in diseases, diagnostics, and therapeutic purposes. 2023).
It has been used for patient clinical letters’ writing (Ali et al., 2023). This
tool has been used in the nucleic acid research (Chatterjee et al. (2023)). 6.1. Diagnostics
The DL-enabled powerful tool is also used in drug discovery and
development (Chakraborty et al., 2023b; Pal et al., 2023b). However, ML and DL are performing several diagnostics activities, including
the use of this powerful tool is increasing very fast daily. clinical and omics-data, medical image detection such as early cancer
Furthermore, there are doubts about the accuracy of ChatGPT’s diagnostics, quick detection of infectious diseases, diagnostics applica
translation findings because the program periodically displays unpre tions in different fields of gastroenterology, cardiology, ophthalmology,
dictability in its responses and occasionally omits or oversimplifies the etc. They can be applied to diagnose pathology-related data. Besides
important details. Depending on ChatGPT’s response could create con these diagnostic activities, ML and DL models are used to analyze the
flict with the in-person counseling recommendation of doctors, which EMR or HER data (Fig. 6) (He et al., 2019).
14
Table 7 Choudhury, 2023).

Different chatbot technologies to study different medical issues.
Sl. Different Medical issue and application Reference 6.4. Hospital administration and regulation
no Technologies
1. Chatbot Mental health support in China (Zhang et al., ML and Dl algorithms can guide the day-to-day activity of hospital
2023) administration and regulation. It also helps preserve and analyze the
2. ChatGPT Support tool for breast tumor (Sorin et al., 2023) EMR or HER data. Finally, it might help in disease monitoring (Fig. 6).
board.
3. Chatbot Mental health self-care discovery (Moilanen et al.,
2023) 6.5. Regulatory activity in health regulatory bodies
4. ChatGPT/GPT-4 Support for intensive care unit (Lu et al., 2023b)
medicine
Regulatory bodies can check and measure all the regulatory activity
5. Chatbot Stress and health-related (Schillings et al.,
parameters 2023) related to health using different moles of ML including DL. Regulatory
6. Chatbot Behavioral health during COVID- (Jackson-Triche bodies such as FDA can first review the product-related application
19 et al., 2023) using those models.
7. Vickybot Support for anxiety-depressive (Anmella et al.,
symptoms and work-related 2023)
burnout in primary care and
7. Future challenges during ML to DL implementation in health
healthcare professionals care
8. Chatbot Improving the well-being of type (Boggiss et al.,
1 diabetes 2023) There are numerous critical challenges for those models, such as data
9. ChatGPT Forthcoming diabetes technology (Huang et al.,
standardization, establishing norms, policy, proper regulatory environ
prediction 2023)
10. Chatbot Support for cognitive behavioral (Wang et al., ment, adequately trained and educated AI-related workforce, financial
therapy 2022) implication, Ethical concern for data securities, implementation chal
lenges, etc. (Fig. 6).
11 Chatbot Mental healthcare and (Schick et al.,
assessment 2022)
12 Chatbot (Otis) Used for health anxiety (Goonesekera and 7.1. Data standardization
management Donkin, 2022)
ML to DL results could be more accurate. Therefore, data standard
ization, model training and testing is crucial before it implements in the
patient’s diagnosis. We already discuss the dataset shift, a crucial
6.2. Therapeutics
problem for training and test data (Finlayson et al., 2021). Recently,
Finlayson et al. described real-time problems of those models’ imple
ML and DL can guide physicians and might help in treatment pro
mentation of a sepsis-alerting model at the University of Michigan
cedures considering the HER. It also helps medical surgeons with ma
Hospital in April 2020 (Finlayson et al., 2021). Kruse et al. have
chine intelligence inspired surgery. It might help data-driven precision
described that big data management and improved decision-making are
medicine. These models also help in pharmacogenomics–based medical
the key challenges in the healthcare sector (Kruse and Goswamy, 2016).
therapy (Fig. 6).
Therefore, it is necessary to have proper standards of ML or DL models
before implementing them into the actual medical field.
6.3. Patients/population health management
7.2. Establishing norms, policy, and the regulatory environment
ML and DL models are trying to generate patient-centric information
for healthy lifestyle information and promotion. These models might be ML will be applied in the real world in medicine. Sometime, the
helpful for the diseases detection in early stage. Simultaneously, these technologies might be data-biased. Therefore, ML or DL technologies
models will help public education about healthcare. The chatbot tech can be implemented considering the risk of illness. The regulatory
nologies like DL inspired ChatGPT are helping people to solve health- bodies must control to ensure effectiveness and safety. Therefore, each
related questions (Fig. 6) (Dave et al., 2023; Shahsavar and country should have norms and policies in those directions to ensure
Fig. 5. The broad role of ML and DL in medicine and healthcare.
15
Fig. 6. The challenges and progress of ML and DL in medicine and healthcare.
effectiveness and safety. Therefore, it is necessary to create a regulatory 8. Conclusion

environment.
The ML and DL technologies have made significant progress in
medicine and healthcare. Ongoing progresses in research from ML to DL
7.3. Need adequately trained and educated workforce
areas are creating the recent paradigm shift in medicine and healthcare.
However, the implementation of these technologies in real-world med
The workforce should be adequately trained and educated before
ical fields is still in its infancy and huge concern. These technologies are
implementation of ML or DL technologies. The curriculum of bachelor’s
in the early stages of development. All recent AI–based research tech
and master’s courses in medical schools must be updated and include the
nologies need to scale up for use. At the same time, all the technology
topics like statistics, computer science, and health informatics (Tang
needs proper validation and implementation. Therefore, the technolo
et al., 2018). All fellowships of residency need to go through dedicated
gies must be scalable and applicable. Similarly, during technological
training programs and projects of informatics. Nurses should also be
implementation and use, there should be collaboration among computer
trained in this direction.
scientists, physicians and data scientists, healthcare providers, and en
gineers, which is essential now. At the same time, new and diversified
7.4. Technology implementation and financial implication algorithms are needed with numerous medical applications and to
improve the quality of diagnostics and therapeutics.
These systems are expensive. Therefore, all the technologies imple The successful integration of ML and DL technologies in the health
mentation needs a considerable amount of cost. At the same time, the care sector requires a multifaceted and flexible approach. The founda
systems need ongoing maintenance, such as software updates, tion of this approach lies in fostering interdisciplinary collaboration
algorithms-related software updates, hardware maintenance, and sub among experts in AI, medicine, data science, and engineering. Estab
stantial financial implications associated with these technologies. lishing regulatory frameworks and standards tailored explicitly to AI in
Therefore, before implementing the technologies, proper budget allo healthcare is crucial, and international cooperation is essential to ensure
cation should be needed. consistency and comprehensiveness. These standards should encompass
important ethical considerations such as data privacy, transparency, and
7.5. Ethical concern for data securities accountability. Rigorous validation protocols should include random
ized controlled trials and assessments using real-world data and
ML technologies need cyber security measures related to patient comparing the performance of these algorithms against existing clinical
safety. The misconduct might take place, and hackers can hack the practices to ensure the reliability of AI algorithms. Comprehensive ed
medical datasets. It should consider that it includes sensitive informa ucation and training programs should be implemented to cultivate talent
tion from real-time patients. Therefore, it is necessary to maintain the and expertise, targeting both healthcare professionals and AI specialists.
patient’s privacy. Design principles that prioritize the patient’s needs should not only
focus on the accuracy of diagnostics and therapeutics but also on
enhancing the overall patient experience, emphasizing empathy and
7.6. Other implementation challenges accessibility. Open-source initiatives can be employed to expedite
research and development. Government and industry support, including
There is a concern for large image sizes because neural networks can funding, incentives, and regulatory cooperation, will be vital in scaling
increase the model complexity and, thereby, the number of pixels. up and widely implementing AI technologies in healthcare. Lastly,
Hence, there is an increased amount of memory required. Another maintaining a culture of continuous learning and improvement is
limitation of the dataset is the risk of noise in some weak-supervision paramount as AI technologies evolve rapidly. Healthcare systems should
setups (Rajpurkar et al., 2022). It is also a need for a technological include flexibility and adaptability, with a commitment to staying at the
bias system when data is collected from different sources.
16
forefront of technological advancements and embracing change to Beam, A.L., Kohane, I.S., 2018. Big Data and Machine Learning in Health Care. J. Am.
Med. Assoc. 319 (13), 1317–1318.
enhance the quality of care. This comprehensive strategy for ML to DL-
Benjamens, S., Dhunnoo, P., et al., 2020. The state of artificial intelligence-based FDA-
based medical technologies will address the current concerns sur approved medical devices and algorithms: an online database. NPJ Digit Med 3, 118.
rounding the implementation of AI in healthcare but also ensure a Bharati, S., et al., 2020. Diagnosis of Polycystic Ovary Syndrome Using Machine Learning
brighter future for the field as it enters a new era propelled by cutting- Algorithms. In: 2020 IEEE Region 10 Symposium (TENSYMP), pp. 1486–1489.
https://doi.org/10.1109/TENSYMP50017.2020.9230932.
edge technologies. Bharati, S., et al., 2023. A Review on Explainable Artificial Intelligence for Healthcare:
Why, How, and When? IEEE Transactions on Artificial Intelligence. https://doi.org/
CRediT authorship contribution statement 10.1109/TAI.2023.3266418.
Bharati, S., et al. (2018). “Breast Cancer Prediction Applying Different Classification
Algorithm with Comparative Analysis using WEKA,” 2018 4th International
Chiranjib Chakraborty: Conceptualization, Data curation, Writing Conference on Electrical Engineering and Information & Communication
– original draft, Writing – review & editing, Supervision, Project Technology (iCEEiCT), Dhaka, Bangladesh, 581-584, doi: 10.1109/
CEEICT.2018.8628084.
administration. Manojit Bhattacharya: Validation, Formal analysis. Biswas, S.S., 2023. “Evaluating Errors and Improving Performance of ChatGPT: A
Soumen Pal: Investigation, Formal analysis. Sang-Soo Lee: Validation, Research Paper. Qeios.
Formal analysis. Bleicher, L.S., van Daelen, T., et al., 2022. Enhanced utility of AI/ML methods during
lead optimization by inclusion of 3D ligand information. Frontiers in Drug Discovery
2, 46.
Declaration of competing interest Boggiss, A., Consedine, N., et al., 2023. Improving the Well-being of Adolescents With
Type 1 Diabetes During the COVID-19 Pandemic: Qualitative Study Exploring
Acceptability and Clinical Usability of a Self-compassion Chatbot. JMIR Diabetes 8,
The authors declare that they have no known competing financial e40641.
interests or personal relationships that could have appeared to influence Cabitza, F., Rasoini, R., et al., 2017. Unintended Consequences of Machine Learning in
the work reported in this paper. Medicine. J. Am. Med. Assoc. 318 (6), 517–518.
Caldarini, G., Jaf, S., et al., 2022. A literature survey of recent advances in chatbots.
Information 13 (1), 41.
Data availability Cammarota G, G, Ianiro G, (2020), Tortora G. Gut microbiome, big data and machine
learning to promote precision medicine for cancer. Nat Rev Gastroenterol Hepatol.
No data was used for the research described in the article. 17(10):635-648. doi: 10.1038/s41575-020-0327-3.
Campanella, G., Hanna, M.G., et al., 2019. Clinical-grade computational pathology using
weakly supervised deep learning on whole slide images. Nat. Med. 25 (8),
Acknowledgement 1301–1309.
Cao, W., et al., 2022. Geometric machine learning: research and applications. Multimed.
Tools Appl. 81, 30545–30597. https://doi.org/10.1007/s11042-022-12683-9.
This study was supported by Hallym University Research Fund and Castiglioni, I., Rundo, L., et al., 2021. AI applications to medical images: From machine
by Basic Science Research Program through the National Research learning to deep learning. Phys. Med. 83, 9–24.
Foundation of Korea (NRF) funded by the Ministry of Education (NRF- Chafai, N., Bonizzi, L., 2023. Emerging applications of machine learning in genomic
medicine and healthcare. Crit Rev Clin Lab Sci.:1–24. https://doi.org/10.1080/
2020R1I1A3074575). 10408363.2023.2259466.
Chahal, D., Byrne, M.F., 2020. A primer on artificial intelligence and its application to
References endoscopy. Gastrointest. Endosc. 92 (4), 813–820 e814.
Chakraborty, C., Bhattacharya, M., et al., 2022. Deep learning research should be
encouraged more and more in different domains of surgery: An open call –
Abidi, S.S.R., Abidi, S.R., 2019. Intelligent health data analytics: A convergence of
Correspondence. Int. J. Surg. 104, 106749.
artificial intelligence and big data. Healthc. Manage. Forum 32 (4), 178–182.
Chakraborty, C., Bhattacharya, M., et al., 2022a. Deep learning research should be
Acs, B., Rimm, D.L., 2018. Not Just Digital Pathology, Intelligent Digital Pathology.
encouraged for diagnosis and treatment of antibiotic resistance of microbial
JAMA Oncol. 4 (3), 403–404.
infections in treatment associated emergencies in hospitals. Int. J. Surg 105, 106857.
Adamson, A.S., Smith, A., 2018. Machine Learning and Health Care Disparities in
Chakraborty, C., Bhattacharya, M., et al., 2022b. Structural Landscape of nsp Coding
Dermatology. JAMA Dermatol. 154 (11), 1247–1248.
Genomic Regions of SARS-CoV-2-ssRNA Genome: A Structural Genomics Approach
Ahmed, Z., K. Mohamed, et al. (2020). “Artificial intelligence with multi-functional
Toward Identification of Druggable Genome, Ligand-Binding Pockets, and Structure-
machine learning platform development for better healthcare and precision
Based Druggability. Mol. Biotechnol. 4, 1–22. https://doi.org/10.1007/s12033-022-
medicine.” Database (Oxford) 2020.
00605-x.
Aisu, N., Miyake, M., et al., 2022. Regulatory-approved deep learning/machine learning-
Chakraborty, C., Bhattacharya, M., et al., 2023a. Artificial intelligence enabled ChatGPT
based medical devices in Japan as of 2020: A systematic review. PLOS Digit Health 1
and large language models in drug target discovery, drug discovery, and
(1), e0000001.
development. Mol. Ther. Nucleic Acids 2023 (33), 866–868. https://doi.org/
Ali, S.R., Dobbs, T.D., et al., 2023. Using ChatGPT to write patient clinic letters. Lancet
10.1016/j.omtn.2023.08.009.
Digit Health 5 (4), e179–e181. https://doi.org/10.1016/S2589-7500(23)00048-1.
Chakraborty, C., Bhattacharya, M., et al., 2023b. Need an AI-Enabled, Next-Generation,
Alizadehsani, R., Roshanzamir, M., et al., 2021. Handling of uncertainty in medical data
Advanced ChatGPT or Large Language Models (LLMs) for Error-Free and Accurate
using machine learning and probability theory techniques: a review of 30 years
Medical Information. Ann. Biomed. Eng. https://doi.org/10.1007/s10439-023-
(1991–2020). Annals of Operations Research 1–42.
03297-9.
Alley, E.C., Khimulya, G., et al., 2019. Unified rational protein engineering with
Chakraborty, C., Pal, S., et al., 2023. Overview of Chatbots with special emphasis on
sequence-based deep representation learning. Nat. Methods 16 (12), 1315–1322.
artificial intelligence-enabled ChatGPT in medical science. Front. Artif. Intell. 6,
Altaf, M.M., 2021. A hybrid deep learning model for breast cancer diagnosis based on
1237704. https://doi.org/10.3389/frai.2023.1237704.
transfer learning and pulse-coupled neural networks. Math. Biosci. Eng. 18 (5),
Char, D.S., Shah, N.H., et al., 2018. Implementing Machine Learning in Health Care –
5029–5046.
Addressing Ethical Challenges. N. Engl. J. Med. 378 (11), 981–983.
Amri, M.M., Hisan, U.K., 2023. Incorporating AI Tools into Medical Education:
Chatterjee, S., Bhattacharya, M., et al., 2023. Can artificial intelligence-strengthened
Harnessing the Benefits of ChatGPT and Dall-E. J. Novel Eng. Sci. Technol. 2 (02),
ChatGPT or other large language models transform nucleic acid research? Mol. Ther.
34–39.
Nucleic Acids 33, 205–207. https://doi.org/10.1016/j.omtn.2023.06.019.
Anmella, G., Sanabra, M., et al., 2023. Vickybot, a Chatbot for Anxiety-Depressive
Chinnadurai S, S, Mahadevan, et al (2023). “Decoding Applications of Artificial
Symptoms and Work-Related Burnout in Primary Care and Health Care
Intelligence in Rheumatology”. Cureus. Sep 28;15(9):e46164. doi: 10.7759/
Professionals: Development, Feasibility, and Potential Effectiveness Studies. J. Med.
cureus.46164.
Internet Res. 25, e43293.
Chiu, Y.C., Chen, H.H., et al., 2020. (2020) Deep learning of pharmacogenomics
Arango-Argoty, G., Garner, E., et al., 2018. DeepARG: a deep learning approach for
resources: moving towards precision oncology. Brief. Bioinform. 21 (6), 2066–2083.
predicting antibiotic resistance genes from metagenomic data. Microbiome 6 (1), 23.
Choi, E., Bahadori, M.T., et al., 2016a. Doctor AI: Predicting Clinical Events via
Ardila, D., Kiraly, A.P., et al., 2019. End-to-end lung cancer screening with three-
Recurrent Neural Networks. JMLR Workshop Conf Proc 56, 301–318.
dimensional deep learning on low-dose chest computed tomography. Nat. Med. 25
Choi, E., Bahadori, M.T., et al., 2016b. Multi-layer representation learning for medical
(6), 954–961.
concepts. In: Proceedings of the 22Nd ACM SIGKDD International Conference on
Ayala Solares, J.R., Diletta Raimondi, F.E., et al., 2020. Deep learning for electronic
Knowledge Discovery and Data Mining, pp. 1495–1504. https://doi.org/10.1145/
health records: A comparative review of multiple deep neural architectures.
2939672.2939823.
J. Biomed. Inform. 101, 103337.
Choi, E., Bahadori, M.T., et al., 2016c. Retain: An interpretable predictive model for
Azencott, C.A., 2018. Machine learning and genomics: precision medicine versus patient
healthcare using reverse time attention mechanism. Adv. Neural Inf. Proces. Syst. 29
privacy. Philos Trans A Math Phys Eng Sci 376 (2128).
https://doi.org/10.48550/arXiv.1608.05745.
Bajaj, S., Gandhi, D., et al., 2023. Potential Applications and Impact of ChatGPT in
Chung, W.Y., Fong, E.M., 2014. Seamless personal health information system in cloud
Radiology. Acad Radiol,:S1076–6332(23), 00460–00469. https://doi.org/10.1016/
computing. Annu Int Conf IEEE Eng Med Biol Soc 2014, 3658–3661.
j.acra.2023.08.039.
17
Cirillo, D., Valencia, A., 2019. Big data analytics for personalized medicine. Curr. Opin. Hersh, W.R., Weiner, M.G., et al., 2013. Caveats for the use of operational electronic
Biotechnol. 58, 161–167. health record data in comparative effectiveness research. Med. Care 51 (8 Suppl 3),
Crown, W.H., 2019. Real-World Evidence, Causal Inference, and Machine Learning. S30–S37.
Value Health 22 (5), 587–592. Hinton, G., 2018. Deep Learning-A Technology With the Potential to Transform Health
Dave, T., Athaluri, S.A., et al., 2023. ChatGPT in medicine: an overview of its Care. J. Am. Med. Assoc. 320 (11), 1101–1102.
applications, advantages, limitations, future prospects, and ethical considerations. Hisan, U.K., Amri, M.M., 2023. ChatGPT and medical education: A double-edged sword.
Front Artif Intell 6, 1169595. J. Pedagogy Educ. Sci. 2 (01), 71–89.
Devaram, S. (2020). “Empathic chatbot: Emotional intelligence for empathic chatbot: Hollon, T.C., Pandian, B., et al., 2020. Near real-time intraoperative brain tumor
Emotional intelligence for mental health well-being.” arXiv preprint arXiv: diagnosis using stimulated Raman histology and deep neural networks. Nat. Med. 26
2012.09130. (1), 52–58.
Devlin, J., M.-W. Chang, et al. (2018). “Bert: Pre-training of deep bidirectional Hu, H., Huang, H., et al., 2023. A wearable cardiac ultrasound imager. Nature 613
transformers for language understanding.” arXiv preprint arXiv:1810.04805. (7945), 667–675.
Dewangan, K.K., Dewangan, D.K., et al., 2022. Breast cancer diagnosis in an early stage Huang, L., Shea, A.L., et al., 2019. Patient clustering improves efficiency of federated
using novel deep learning with hybrid optimization technique. Multimed. Tools machine learning to predict mortality and hospital stay time using distributed
Appl. 81 (10), 13935–13960. electronic medical records. J. Biomed. Inform. 99, 103291.
Dias, P., Cardoso, M., et al., 2022. Remote Patient Monitoring Systems based on Huang, J., Yeung, A.M., et al., 2023. Using ChatGPT to Predict the Future of Diabetes
Conversational Agents for Health Data Collection. HEALTHINF. https://doi.org/ Technology. J. Diabetes Sci. Technol. 17 (3), 853–854.
10.5220/0011011000003123. Hulsen, T., Jamuar, S.S., et al., 2019. From Big Data to Precision Medicine. Front Med
Dicuonzo, G., Galeone, G., et al., 2022. Towards the Use of Big Data in Healthcare: A (lausanne) 6, 34.
Literature Review. Healthcare (basel) 10 (7). Ibtisum, S.,Comparative Study on Different Big Data Tools 2020. M.Sc., 1-93. thesis.
Dinh, A., Miertschin, S., et al., 2019. A data-driven approach to predicting diabetes and https://hdl.handle.net/10365/31657.
cardiovascular disease with machine learning. BMC Med. Inf. Decis. Making 19 (1), Izci, H., Macq, G., et al., 2023. Machine Learning Algorithm to Estimate Distant Breast
211. Cancer Recurrence at the Population Level with Administrative Data. Clin.
Ebrahimian, S., Kalra, M.K., et al., 2022. FDA-regulated AI Algorithms: Trends, Epidemiol. 15, 559–568.
Strengths, and Gaps of Validation Studies. Acad. Radiol. 29 (4), 559–566. Jaamour, A., Myles, C., et al., 2023. A divide and conquer approach to maximise deep
Emmert-Streib, F., Dehmer, M., 2018. A machine learning perspective on Personalized learning mammography classification accuracies. PLoS One 18 (5), e0280841.
Medicine: an automized, comprehensive knowledge base with ontology for pattern Jackson, H.W., Fischer, J.R., et al., 2020. The single-cell pathology landscape of breast
recognition. Machine Learning and Knowledge Extraction 1 (1), 149–156. cancer. Nature 578 (7796), 615–620.
Esteva, A., Kuprel, B., et al., 2017. Dermatologist-level classification of skin cancer with Jackson-Triche, M., Vetal, D., et al., 2023. Meeting the Behavioral Health Needs of
deep neural networks. Nature 542 (7639), 115–118. Health Care Workers During COVID-19 by Leveraging Chatbot Technology. J. Med.
Fadhil, A. and G. Schiavo (2019). “Designing for health chatbots.” arXiv preprint arXiv: Internet Res.
1902.09022. Jacob, C., Lindeque, J., et al., 2023. Assessing the Quality and Impact of eHealth Tools:
A. Fadhil G. Schiavo et al. “CoachAI: A conversational agent assisted health coaching Systematic Literature Review and Narrative Synthesis. JMIR Hum. Factors 10,
platform”. arXiv preprint arXiv:1904.11961 2019. e45143.
Finlayson, S.G., Subbaswamy, A., et al., 2021. The Clinician and Dataset Shift in Artificial Jassar, S., Adams, S.J., et al., 2022. The future of artificial intelligence in medicine:
Intelligence. N. Engl. J. Med. 385 (3), 283–286. Medical-legal considerations for health leaders. Healthc. Manage. Forum 35 (3),
Fletcher, A.J., Lapidaire, W., et al., 2021. Machine Learning Augmented 185–189.
Echocardiography for Diastolic Function Assessment. Front Cardiovasc Med 8, Jia, J. (2003). “The study of the application of a keywords-based chatbot system on the
711611. teaching of foreign languages.” arXiv preprint cs/0310018.
Fu, Q., Zhang, Y., et al., 2021. Rapid identification of the resistance of urinary tract Jiang, F., Jiang, Y., et al., 2017. Artificial intelligence in healthcare: past, present and
pathogenic bacteria using deep learning-based spectroscopic analysis. Anal. Bioanal. future. Stroke Vasc Neurol 2 (4), 230–243.
Chem. 413 (30), 7401–7410. Jimenez-Carretero, D., Abrishami, V., et al., 2018. Tox_(R)CNN: Deep learning-based
Gainza, P., Sverrisson, F., et al., 2020. Deciphering interaction fingerprints from protein nuclei profiling tool for drug toxicity screening. PLoS Comput. Biol. 14 (11),
molecular surfaces using geometric deep learning. Nat. Methods 17 (2), 184–192. e1006238.
Gaudelet, T., et al., 2021. Utilizing graph machine learning within drug discovery and Jin, X., et al., 2023. Artificial intelligence biosensors for continuous glucose monitoring.
development. Brief. Bioinform. 22(6):bbab159 https://doi.org/10.1093/bib/ Interdisciplinary Materials 2, 290–307. https://doi.org/10.1002/idm2.12069.
bbab159. Jin, S., et al. (2023). “Deep Learning in COVID-19 Diagnosis, Prognosis and Treatment
Ghorbani, A., Ouyang, D., et al., 2020. Deep learning interpretation of echocardiograms. Selection.” Mathematics 11, no. 6 (2023): 1279. Doi: 10.3390/math11061279.
NPJ Digit Med 3, 10. https://doi.org/10.1038/s41746-019-0216-8. Jones, L.D., Golan, D., et al., 2018. Artificial intelligence, machine learning and the
Gilson, A., Safranek, C.W., et al., 2023. How Does ChatGPT Perform on the United States evolution of healthcare: A bright future or cause for concern? Bone Joint Res 7 (3),
Medical Licensing Examination? The Implications of Large Language Models for 223–225.
Medical Education and Knowledge Assessment. JMIR Med Educ 9, e45312. Joublin, F., A. Ceravola, et al. (2023). “A Glimpse in ChatGPT Capabilities and its impact
Goecks, J., Jalili, V., et al., 2020. How Machine Learning Will Transform Biomedicine. for AI research.” arXiv preprint arXiv:2305.06087.
Cell 181 (1), 92–101. Jovanović, M., Baez, M., et al., 2020. Chatbots as conversational healthcare services.
Gong, D., Wu, L., et al., 2020. Detection of colorectal adenomas with a real-time IEEE Internet Comput. 25 (3), 44–51.
computer-aided system (ENDOANGEL): a randomised controlled study. Lancet Kang, I.-A., Ngnamsie Njimbouom, S., et al., 2022. DCP: prediction of dental caries using
Gastroenterol. Hepatol. 5 (4), 352–361. machine learning in personalized medicine. Appl. Sci. 12 (6), 3043.
Goonesekera, Y., Donkin, L., 2022. A Cognitive Behavioral Therapy Chatbot (Otis) for Kaul, V., Enslin, S., et al., 2020. History of artificial intelligence in medicine.
Health Anxiety Management: Mixed Methods Pilot Study. JMIR Form Res 6 (10), Gastrointest. Endosc. 92 (4), 807–812.
e37877. Kaul, D., et al. (2022). Deep Learning in Healthcare. In: Acharjya, D.P., Mitra, A., Zaman,
Green, H., Durrant, J.D., 2021. DeepFrag: An Open-Source Browser App for Deep- N. (eds) Deep Learning in Data Analytics. Studies in Big Data, vol 91. Springer,
Learning Lead Optimization. J. Chem. Inf. Model. 61 (6), 2523–2529. Cham. Doi: 10.1007/978-3-030-75855-4_6.
Greener, J.G., Kandathil, S.M., et al., 2019. Deep learning extends de novo protein Kilic, A., 2020. Artificial Intelligence and Machine Learning in Cardiovascular Health
modelling coverage of genomes using iteratively predicted structural constraints. Care. Ann. Thorac. Surg. 109 (5), 1323–1329. https://doi.org/10.1016/j.
Nat. Commun. 10 (1), 3977. athoracsur.2019.09.042.
Gruson, D., Helleputte, T., et al., 2019. Data science, artificial intelligence, and machine Kourou, K., Exarchos, T.P., et al., 2015. Machine learning applications in cancer
learning: Opportunities for laboratory medicine and the value of positive regulation. prognosis and prediction. Comput. Struct. Biotechnol. J. 13, 8–17.
Clin. Biochem. 69, 1–7. https://doi.org/10.1016/j.clinbiochem.2019.04.013. Kumar, D., et al., 2023a. Exploring the Transformative Role of Artificial Intelligence and
Gulshan, V., Peng, L., et al., 2016. Development and Validation of a Deep Learning Metaverse in Education: A Comprehensive Review. Metaverse Basic and Applied
Algorithm for Detection of Diabetic Retinopathy in Retinal Fundus Photographs. Research. https://doi.org/10.56294/mr202355.
J. Am. Med. Assoc. 316 (22), 2402–2410. Kumar, N., Acharya, V., 2022. Machine intelligence-driven framework for optimized hit
Gussow, A.B., Park, A.E., et al., 2020. Machine-learning approach expands the repertoire selection in virtual screening. J Cheminform 14 (1), 48.
of anti-CRISPR protein families. Nat. Commun. 11 (1), 3784. Kumar, K., Chaudhury, K., et al., 2023b. “Future of Machine Learning (ML) and Deep
Hammad, M., Pławiak, P., 2023. Enhanced Deep Learning Approach for Accurate Eczema Learning (DL) in Healthcare Monitoring System,” in Machine Learning Algorithms
and Psoriasis Skin Detection. Sensors (basel). 23 (16), 7295. for Signal and Image Processing. IEEE 293–313. https://doi.org/10.1002/
Hashana, A.J., Brundha, P., et al., 2023. Deep Learning in ChatGPT-A Survey. In: 2023 9781119861850.ch17.
7th International Conference on Trends in Electronics and Informatics (ICOEI), IEEE. Kurokowa, M., Lynch, K., et al., 1987. Effects of growth factors on an intestinal epithelial
ISBN:979-8-3503-9729-1. https://doi.org/10.1109/ICOEI56765.2023.10125852. cell line: transforming growth factor beta inhibits proliferation and stimulates
Hassan, M., Awan, F.M., et al., 2022. Innovations in Genomics and Big Data Analytics for differentiation. Biochem. Biophys. Res. Commun. 142 (3), 775–782.
Personalized Medicine and Health Care: A Review. Int. J. Mol. Sci. 23 (9), 4645. LeCun, Y., Bengio, Y., et al., 2015. Deep learning. Nature 521 (7553), 436–444.
https://doi.org/10.3390/ijms23094645. Lee, C.Y., Chen, Y.P., 2021. Prediction of drug adverse events using deep learning in
Haug, C.J., Drazen, J.M., 2023. Artificial Intelligence and Machine Learning in Clinical pharmaceutical discovery. Brief. Bioinform. 22 (2), 1884–1901.
Medicine, 2023. N. Engl. J. Med. 388 (13), 1201–1208. Leiserson, C.E., Thompson, N.C., et al., 2020. There’s plenty of room at the Top: What
He, J., Baxter, S.L., et al., 2019. The practical implementation of artificial intelligence will drive computer performance after Moore’s law? Science 368 (6495).
technologies in medicine. Nat. Med. 25 (1), 30–36. Leite, A.F., Vasconcelos, K.F., 2020. Radiomics and Machine Learning in Oral Healthcare.
Proteomics Clin. Appl. 14 (3), e1900040.
18
Li, Y., Xu, Z., et al., 2021. HMD-ARG: hierarchical multi-task deep learning for use of the artificial intelligence (AI)-based device for a faster way of drug discovery
annotating antibiotic resistance genes. Microbiome 9 (1), 40. and development. Int. J. Surg. https://doi.org/10.1097/JS9.0000000000000719.
Krittanawong, C., Johnson, K.W., et al., 2019. Deep learning for cardiovascular medicine: Pandey, A., Kagiyama, N., et al., 2021. Deep-Learning Models for the Echocardiographic
a practical primer. Eur. Heart J. 40 (25), 2058–2073. Assessment of Diastolic Dysfunction. J. Am. Coll. Cardiol. Img. 14 (10), 1887–1900.
Kruse, C.S., Goswamy, R., et al., 2016. Challenges and Opportunities of Big Data in Peng, Y, and Lu, Z (2017). “Deep learning for extracting protein-protein interactions
Health Care: A Systematic Review. JMIR Med. Inform. 4 (4), e38. from biomedical literature.” arXiv preprint arXiv:1706.01556. Doi: 10.48550/
Li, J., A. Dada, et al. (2023). “ChatGPT in Healthcare: A Taxonomy and Systematic arXiv.1706.01556.
Review.” medRxiv: 2023.2003. 2030.23287899. Pham, T., Tran, T., et al., 2017. Predicting healthcare trajectories from medical records:
Liu, Y., Jain, A., et al., 2020a. A deep learning system for differential diagnosis of skin A deep learning approach. J. Biomed. Inform. 69, 218–229.
diseases. Nat. Med. 26 (6), 900–908. https://doi.org/10.1038/s41591-020-0842-3. Phatak, A.A., et al., 2021. Artificial Intelligence Based Body Sensor Network Framework-
Liu, H., Li, L., et al., 2019. Development and Validation of a Deep Learning System to Narrative Review: Proposing an End-to-End Framework using Wearable Sensors,
Detect Glaucomatous Optic Neuropathy Using Fundus Photographs. JAMA Real-Time Location Systems and Artificial Intelligence/Machine Learning
Ophthalmol 137 (12), 1353–1360. Algorithms for Data Collection, Data Mining and Knowledge Discovery in Sports and
Liu, T., Siegel, E., et al., 2022. Deep Learning and Medical Image Analysis for COVID-19 Healthcare. Sports Med Open 7 (1), 79. https://doi.org/10.1186/s40798-021-
Diagnosis and Prediction. Annu. Rev. Biomed. Eng. 24, 179–201. 00372-0.
Liu, P., Wang, Z., et al., 2023. A scoping review of the clinical application of machine Pournaras, E., 2023. “Science in the Era of ChatGPT, Large Language Models and AI.
learning in data-driven population segmentation analysis. J. Am. Med. Inform. Challenges for Research Ethics Review and How to Respond”. arXiv preprint arXiv:
Assoc. 30 (9), 1573–1582. 2305.15299.
Liu, X., Yang, Y., et al., 2020b. SeqCor: correct the effect of guide RNA sequences in Qureshi, R., et al., 2023. Artificial Intelligence and Biosensors in Healthcare and its
clustered regularly interspaced short palindromic repeats/Cas9 screening by Clinical Relevance: A Review. IEEE Access 11, 61600–61620. https://doi.org/
machine learning algorithm. J. Genet. Genomics 47 (11), 672–680. 10.1109/ACCESS.2023.3285596.
Liu, Z., J. Du, et al. (2019). “DeepScreening: a deep learning-based screening web server Raghupathi, W., Raghupathi, V., 2014. Big data analytics in healthcare: promise and
for accelerating drug discovery.” Database (Oxford) 2019: baz104. Doi: 10.1093/ potential. Health Inf Sci Syst 2, 3. https://doi.org/10.1186/2047-2501-2-3.
database/baz104. Rahmani, A.M., et al., 2021. “Machine learning (ML) in medicine: Review, applications,
Lu, T., et al., 2023a. Biocompatible and Long-Term Monitoring Strategies of Wearable, and challenges. Mathematics” 9 (22), 2970. https://doi.org/10.3390/math9222970.
Ingestible and Implantable Biosensors: Reform the Next Generation Healthcare. Rajkomar, A., Dean, J., et al., 2019. Machine Learning in Medicine. N. Engl. J. Med. 380
Sensors (basel). 23 (6), 2991. https://doi.org/10.3390/s23062991. (14), 1347–1358.
Lu, Y., Wu, H., et al., 2023b. Artificial Intelligence in Intensive Care Medicine: Toward a Rajkomar, A., Oren, E., et al., 2018. Scalable and accurate deep learning with electronic
ChatGPT/GPT-4 Way? Ann. Biomed. Eng. 9, 1898–1903. health records. NPJ Digit Med 1, 18. https://doi.org/10.1038/s41746-018-0029-1.
Lyu, Q., Tan, J., et al., 2023. Translating radiology reports into plain language using Rajpurkar, P., Chen, E., et al., 2022. AI in health and medicine. Nat. Med. 28 (1), 31–38.
ChatGPT and GPT-4 with prompt learning: results, limitations, and potential. Vis Rajpurkar, P., Irvin, J., et al., 2018. Deep learning for chest radiograph diagnosis: A
Comput Ind Biomed Art 6 (1), 9. retrospective comparison of the CheXNeXt algorithm to practicing radiologists. PLoS
Marshall, D.A., Burgos-Liz, L., et al., 2016. Transforming Healthcare Delivery: Med. 15 (11), e1002686.
Integrating Dynamic Simulation Modelling and Big Data in Health Economics and Rajput, A. and A. Subasi (2023). Lung cancer detection from histopathological lung tissue
Outcomes Research. Pharmacoeconomics 34 (2), 115–126. images using deep learning. In book Applications of Artificial Intelligence in Medical
Matsushita F.Y, V.L.J Krebs, (2022). “Artificial intelligence and machine learning in Imaging, Elsevier: 51-74. ISBN 9780443184505, Doi: 10.1016/B978-0-443-18450-
pediatrics and neonatology healthcare”. Rev Assoc Med Bras (1992).;68(6):745-750. 5.00008-6.
May, M., 2021. Eight ways machine learning is assisting medicine. Nat. Med. 27 (1), 2–3. Ramamurthi A, C Are, et al (2023) From ChatGPT to Treatment: the Future of AI and
Megahed, F. M., Y.-J. Chen, et al. (2023). “How generative ai models such as chatgpt can Large Language Models in Surgical Oncology. Indian J Surg Oncol. 2023 Sep;14(3):
be (mis) used in spc practice, education, and research? An exploratory study.” arXiv 537-539. doi: 10.1007/s13193-023-01836-3.
preprint arXiv:2302.10916. M.I. Razzak S. Naz et al. Deep Learning for Medical Image Processing: Overview,
Milea, D., Najjar, R.P., et al., 2020. Artificial Intelligence to Detect Papilledema from Challenges and the Future N. Dey A. Ashour S. Borra Classification in BioApps
Ocular Fundus Photographs. N. Engl. J. Med. 382 (18), 1687–1695. Lecture Notes in Computational Vision and Biomechanics 26 2018 Springer Cham
Millman, R.P., Fogel, B.S., et al., 1989. Depression as a manifestation of obstructive sleep 10.1007/978-3-319-65981-7_12.
apnea: reversal with nasal continuous positive airway pressure. J. Clin. Psychiatry 50 Ristevski, B., Chen, M., 2018. Big Data Analytics in Medicine and Healthcare. J. Integr.
(9), 348–351. Bioinform. 15 (3), 20170030.
Miotto, R., Li, L., et al., 2016. Deep Patient: An Unsupervised Representation to Predict Rose, S., 2018. Machine Learning for Prediction in Electronic Health Data. JAMA Netw.
the Future of Patients from the Electronic Health Records. Sci. Rep. 6, 26094. Open 1 (4), e181404.
Mitchell, E. G., R. Maimone, et al. (2021). “Automated vs. Human Health Coaching: Rubinger, L., et al., 2023. Machine learning and artificial intelligence in research and
Exploring Participant and Practitioner Experiences.” Proc ACM Hum Comput healthcare. Injury 54, S69–S73. https://doi.org/10.1016/j.injury.2022.01.046.
Interact 5(CSCW1). Doi: 10.1145/3449173. Sabour, S., Zhang, W., et al., 2023. A chatbot for mental health support: exploring the
Moilanen, J., van Berkel, N., et al., 2023. Supporting mental health self-care discovery impact of Emohaa on reducing mental distress in China. Front. Digit. Health 5,
through a chatbot. Front Digit Health 5, 1034724. 1133987.
Mondal, M.R.H., Bharati, S., et al., 2023. Deep Learning and Federated Learning for Saheb, T., Saheb, M., 2019. Analyzing and Visualizing Knowledge Structures of Health
Screening COVID-19. A Review. Biomedinformatics 3 (3), 691–713. https://doi.org/ Informatics from 1974 to 2018: A Bibliometric and Social Network Analysis. Healthc
10.3390/biomedinformatics3030045. Inform Res 25 (2), 61–72.
Najafabadi, M.M., Villanustre, F., et al., 2015. Deep learning applications and challenges Sarker, B. et al. (2023). “AI, IoMT and Blockchain in Healthcare”. Journal of Trends in
in big data analytics. Journal of Big Data 2 (1), 1–21. Computer Science and Smart Technology, 5(1), 30-50, 2023. Doi:10.36548/
Nayak, L., Ray, I., et al., 2016. Precision medicine with electronic medical records: from jtcsst.2023.1.003.
the patients and for the patients. Ann Transl Med 4 (Suppl 1), S61. https://doi.org/ Schick, A., Feine, J., et al., 2022. Validity of Chatbot Use for Mental Health Assessment:
10.21037/atm.2016.10.40. Experimental Study. JMIR Mhealth Uhealth 10 (10), e28082.
Naylor, C.D., 2018. On the Prospects for a (Deep) Learning Health Care System. J. Am. Schillings, C., Meissner, D., et al., 2023. A chatbot-based intervention with ELME to
Med. Assoc. 320 (11), 1099–1100. improve stress and health-related parameters in a stressed sample: Study protocol of
Ng, M.-F., Zhao, J., et al., 2020. Predicting the state of charge and health of batteries a randomised controlled trial. Front Digit Health 5, 1046202.
using data-driven machine learning. Nature Machine Intelligence 2 (3), 161–170. Sedaghat, S., 2023. Early applications of ChatGPT in medical practice, education and
Ngiam, K.Y., Khor, I.W., 2019. Big data and machine learning algorithms for health-care research. Clin. Med. (Lond.) 23 (3), 278–279.
delivery. Lancet Oncol. 20 (5), e262–e273. Senior, A.W., Evans, R., et al., 2020. Improved protein structure prediction using
Nguyen, P., Tran, T., et al., 2017. $\mathtt Deepr$: A Convolutional Net for Medical potentials from deep learning. Nature 577 (7792), 706–710.
Records. IEEE J. Biomed. Health Inform. 21 (1), 22–30. Shahsavar, Y., Choudhury, A., 2023. User Intentions to Use ChatGPT for Self-Diagnosis
Nicol, G., Wang, R., et al., 2022. Chatbot-Delivered Cognitive Behavioral Therapy in and Health-Related Purposes: Cross-sectional Survey Study. JMIR Hum. Factors 10,
Adolescents With Depression and Anxiety During the COVID-19 Pandemic: e47564.
Feasibility and Acceptability Study. JMIR Form Res 6 (11), e40242. Shamshirband, S., et al (2021).” A review on deep learning approaches in healthcare
Obermeyer, Z., Emanuel, E.J., 2016. Predicting the Future – Big Data, Machine Learning, systems: Taxonomies, challenges, and open issues”. J Biomed Inform. Jan;113:
and Clinical Medicine. N. Engl. J. Med. 375 (13), 1216–1219. 103627. doi: 10.1016/j.jbi.2020.103627.
Obstfeld, A.E., 2023. Hematology and Machine Learning. J. Appl. Lab. Med. 8 (1), Shen, L., Margolies, L.R., et al., 2019. Deep Learning to Improve Breast Cancer Detection
129–144. on Screening Mammography. Sci. Rep. 9 (1), 12495.
Oh, S.H., Lee, S.J., et al., 2022. Effective data-driven precision medicine by cluster- Shen, D., Wu, G., et al., 2017. Deep Learning in Medical Image Analysis. Annu. Rev.
applied deep reinforcement learning. Knowl.-Based Syst. 256, 109877. Biomed. Eng. 21 (19), 221–248. https://doi.org/10.1146/annurev-bioeng-071516-
Ostberg, N.P., Zafar, M.A., 2021. Machine learning: principles and applications for 044442.
thoracic surgery. Eur. J. Cardiothorac. Surg. 60 (2), 213–221. https://doi.org/ Shimazaki, A., Ueda, D., et al., 2022. Deep learning-based algorithm for lung cancer
10.1093/ejcts/ezab095. detection on chest radiographs using the segmentation method. Sci. Rep. 12 (1), 727.
Pal, S., Bhattacharya, M., et al., 2023a. A Domain-Specific Next-Generation Large Sitapati, A., Kim, H., et al., 2017. Integrated precision medicine: the role of electronic
Language Model (LLM) or ChatGPT is Required for Biomedical Engineering and health records in delivering personalized treatment. Wiley Interdiscip. Rev. Syst.
Research. Ann. Biomed. Eng. 10 https://doi.org/10.1007/s10439-023-03306-x. Biol. Med. 9 (3).
Pal, S., Bhattacharya, M., et al., 2023b. ChatGPT or LLM in next-generation drug
discovery and development: Pharmaceutical and biotechnology companies can make
19
Soleymani, F., Paquet, E., et al., 2022. Protein-protein interaction prediction with deep Ward, M.D., Zimmerman, M.I., et al., 2021. Deep learning the structural determinants of
learning: A comprehensive review. Comput. Struct. Biotechnol. J. 20, 5316–5341. protein biochemical properties by comparing structural ensembles with DiffNets.
https://doi.org/10.1016/j.csbj.2022.08.070. Nat. Commun. 12 (1), 3023.
Sorin, V., Klang, E., et al., 2023. Large language model (ChatGPT) as a support tool for Weber, G.M., Mandl, K.D., et al., 2014. Finding the missing link for big biomedical data.
breast tumor board. npj Breast Cancer 9 (1), 44. J. Am. Med. Assoc. 311 (24), 2479–2480.
Stein, J.D., Rahman, M., et al., 2019. Evaluation of an Algorithm for Identifying Ocular Wong, D., Yip, S., 2018. Machine learning classifies cancer. Nature 555 (7697), 446–447.
Conditions in Electronic Health Record Data. JAMA Ophthalmol 137 (5), 491–497. Woodman, R.J., Mangoni, AA., 2023. A comprehensive review of machine learning
Stokes, J.M., Yang, K., et al., 2020. A Deep Learning Approach to Antibiotic Discovery. algorithms and their application in geriatric medicine: present and future. Aging
Cell 181 (2), 475–483. Clin. Exp. Res. 35 (11), 2363–2397.
Subbaswamy, A., Saria, S., 2020. From development to deployment: dataset shift, Wu, X., Liu, C., et al., 2023. Internet of things-enabled real-time health monitoring
causality, and shift-stable models in health AI. Biostatistics 21 (2), 345–352. system using deep learning. Neural Comput. & Applic. 35, 14565–14576. https://
Suganyadevi, S., Seethalakshmi, V., et al., 2022. A review on deep learning in medical doi.org/10.1007/s00521-021-06440-6.
image analysis. Int. J. Multimedia Inform. Retrieval 11 (1), 19–38. https://doi.org/ Wu, N., Phang, J., et al., 2020. Deep Neural Networks Improve Radiologists’
10.1007/s13735-021-00218-1. Performance in Breast Cancer Screening. IEEE Trans. Med. Imaging 39 (4),
Tang, A., Tam, R., et al., 2018. Canadian Association of Radiologists White Paper on 1184–1194.
Artificial Intelligence in Radiology. Can. Assoc. Radiol. J. 69 (2), 120–135. Xiao, Z., T. W. Li, et al. (2023). Inform the Uninformed: Improving Online Informed
Ting D.S.W, L.R Pasquale, (2019). Artificial intelligence and deep learning in Consent Reading with an AI-Powered Chatbot. Proceedings of the 2023 CHI
ophthalmology. Br J Ophthalmol. 2019 Feb;103(2):167-175. Tran, T., T. D. Nguyen, Conference on Human Factors in Computing Systems. Doi: 10.48550/
et al. (2015). “Learning vector representation of medical objects via EMR-driven arXiv.2302.00832.
nonnegative restricted Boltzmann machines (eNRBM).” J Biomed Inform 54: 96-105. Yang, Y.J., Bang, C.S., 2019. Application of artificial intelligence in gastroenterology.
Tran, K.A., Kondrashova, O., et al., 2021. Deep learning in cancer diagnosis, prognosis World J. Gastroenterol. 25 (14), 1666–1683.
and treatment selection. Genome Med. 13 (1), 152. Yasnitsky, L.N., 2020. Artificial Intelligence and Medicine: History, Current State, and
Uche-Anya, E., Anyane-Yeboa, A., et al., 2022. Artificial intelligence in gastroenterology Forecasts for the Future. Curr. Hypertens. Rev. 16 (3), 210–215.
and hepatology: how to advance clinical practice while ensuring health equity. Gut Yuan, Y., Meng, M.Q., 2017. Deep learning for polyp recognition in wireless capsule
71 (9), 1909–1915. https://doi.org/10.1136/gutjnl-2021-326271. endoscopy images. Med. Phys. 44 (4), 1379–1389.
Urban, G., Tripathi, P., et al., 2018. Deep Learning Localizes and Identifies Polyps in Real Zhang, Z., Lin, X., et al., 2023. A hybrid algorithm for clinical decision support in
Time With 96% Accuracy in Screening Colonoscopy. Gastroenterology 155 (4), precision medicine based on machine learning. BMC Bioinf. 24 (1), 3.
1069–1078 e1068. Zhao, Z., Yang, Z., et al., 2016. A protein-protein interaction extraction approach based
Vornholt, P., De Choudhury, M., 2021. Understanding the Role of Social Media-Based on deep neural network. Int. J. Data Min. Bioinform. 15 (2), 145–164. https://doi.
Mental Health Support Among College Students: Survey and Semistructured org/10.1504/IJDMB.2016.076534.
Interviews. JMIR Ment Health 8 (7), e24512. Zhao, Q., Zhu, J., et al., 2023. Chinese diabetes datasets for data-driven machine
Wang, H., Czerminski, R., et al., 2021. Neural networks and deep learning. Emerald learning. Sci. Data 10 (1), 35.
Publishing Limited, The machine age of customer insight, pp. 91–101. Zhavoronkov, A., Ivanenkov, Y.A., et al., 2019. Deep learning enables rapid
Wang, P., Liu, X., et al., 2020. Effect of a deep-learning computer-aided detection system identification of potent DDR1 kinase inhibitors. Nat. Biotechnol. 37 (9), 1038–1040.
on adenoma detection during colonoscopy (CADe-DB trial): a double-blind Zhou, S. Kevin, H.G et al (2021). “A Review of Deep Learning in Medical Imaging:
randomised study. Lancet Gastroenterol. Hepatol. 5 (4), 343–351. Imaging Traits, Technology Trends, Case Studies With Progress Highlights, and
Wang, F., Preininger, A., 2019. AI in Health: State of the Art, Challenges, and Future Future Promises,” in Proceedings of the IEEE, 109, 5, 820-838,, doi: 10.1109/
Directions. Yearb. Med. Inform. 28 (1), 16–26. JPROC.2021.3054390.
Wang, M., Wei, Z., et al., 2022. Deep learning model for multi-classification of infectious Zillmer, T.W., 1986. Stop-loss insurance can reduce employers’ risks. Top. Health Care
diseases from unstructured electronic medical records. BMC Med. Inf. Decis. Making Financ. 12 (4), 68–73.
22 (1), 41. Zvarikova, K., Horak, J., et al., 2022. Machine and Deep Learning Algorithms, Computer
Wang, D., Zhang, C., et al., 2019. Optimized CRISPR guide RNA design for two high- Vision Technologies, and Internet of Things-based Healthcare Monitoring Systems in
fidelity Cas9 variants by deep learning. Nat. Commun. 10 (1), 4284. COVID-19 Prevention, Testing, Detection, and Treatment. Am. J. Med. Res. 9 (1),
145–160. https://doi.org/10.22381/ajmr91202210.
20

From-machine-learning-to-deep-learning--Advances-of-the_2024_Current-Researc

Uploaded by

Copyright:

Available Formats

From-machine-learning-to-deep-learning--Advances-of-the_2024_Current-Researc

Uploaded by

Document Information

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

From-machine-learning-to-deep-learning--Advances-of-the_2024_Current-Researc

Uploaded by

Copyright:

Available Formats

Current Research in Biotechnology 7 (2024) 100164

Contents lists available at ScienceDirect

Current Research in Biotechnology

From machine learning to deep learning: Advances of the recent

Table 2 (continued ) Table 3

Table 3 (continued ) Table 4

Table 5 Table 5 (continued )

Table 5 (continued ) Table 5 (continued )

personalized clinical decision conditions,

1. Deepr Convolutional 300,000 It use to predict (Nguyen

Table 7 Choudhury, 2023).

Fig. 5. The broad role of ML and DL in medicine and healthcare.

Fig. 6. The challenges and progress of ML and DL in medicine and healthcare.

effectiveness and safety. Therefore, it is necessary to create a regulatory 8. Conclusion

You might also like