Volume 9, Issue 10, October– 2024 International Journal of Innovative Science and Research Technology

ISSN No:-2456-2165 https://doi.org/10.38124/ijisrt/IJISRT24OCT089

AI in Healthcare: Bridging the Gap between

Research and Clinical Implementation
Srinivas Lanka1 Pavithra Madala2
Consultant - Clinical Research, University of North Carolina at Chapel Hill
Data Science & Data Engineering

Abstract:- Artificial intelligence (AI) has the potential to Things (IoT), cloud computing, and wearables, to improve
revolutionize healthcare by enhancing diagnostic patient experience, reduce costs, and enhance better health
accuracy, reducing administrative burdens, and outcomes. (8) They aim to process and analyze large amounts
providing personalized treatment. However, the slow of digital medical data accurately and efficiently to shift
adoption of AI in healthcare is due to obstacles associated towards a value-based care model where personalized patient
with ethical considerations, data management, care is prioritized. (9,10)
regulations, and technological capabilities. The results of
our study highlight specific challenges related to ethics, Machine learning (ML) is a fundamental element of
technology, regulatory, social, economic, and workforce artificial intelligence (AI) that comprises, models capable of
barriers that affect the implementation of AI in iterative learning. ML is becoming increasingly common in
healthcare. We aim to improve current knowledge by all major sectors, including healthcare. (11–13) It has
providing a more comprehensive understanding, by significantly improved clinical data interpretation in
bridging the gap, and addressing the barriers to radiology, pathology, and dermatology, thanks to
implement AI in the healthcare sector. convolutional neural networks, standardized data formats,
and extensive repositories. (14)
Keywords:- Artificial Intelligence, Implementation Gap,
Machine Learning, Barriers. Today, artificial intelligence (AI) is omnipresent,
significantly more advanced, and user-friendly than it was
I. INTRODUCTION two decades ago and has become a reliable part of our daily
lives. Recently, there has been a noticeable increase in the
Artificial intelligence (AI) has had a significant impact implementation of AI in healthcare services. In AI,
on patient care, medical diagnosis, and treatment outcomes. technological implementation typically involves developing
It can analyze data with/without human intervention. Its software components to implementing an algorithm. (5)
integration into the healthcare industry brought about a
transformative paradigm shift. (1) Incorporating new and We should also acknowledge that only a fraction of
novel technological advances into clinical practice has led to these algorithms is being utilized in practical clinical settings.
many exciting medical developments over the last five Even a few esteemed medical centers with advanced
decades. Among current areas of research, Artificial technological capabilities are not embracing AI in their daily
Intelligence (AI) stands out for capturing attention and workflows. A recent assessment of Deep Learning (DL)
imagination, with the potential to revolutionize every field. applications using health record data emphasized the critical
(2) AI systems offer a smart solution to reduce the workload need to shift focus towards effectively implementing these
of clinical staff within increasingly saturated healthcare models to have a direct, positive impact on clinical practice.
systems. It is important to note that AI goes beyond simple The algorithms trained on historical data and published in the
data management, providing direct suggestions and literature are not implemented in practical use. This research-
recommendations that shape the clinical decision-making implementation disconnect is a major reason AI is not
process. (3–5) deployed more in clinical settings. To make AI a healthcare
reality, we need to bridge this gap. (15)
Artificial intelligence (AI) has seen remarkable growth
in recent decades, expanding its capabilities and applications. Multiple studies have looked at AI implementation in
In the summer 1956, McCarthy, and colleagues (6) healthcare. Ali et al. (16) reviewed the integration of AI in
introduced the concept of AI. Since then, rapid advancements health systems and the challenges in healthcare professionals'
in computational power, internet connectivity, digitalization, acceptance of AI. Petersson et al. (17) discussed AI adoption
and cumulative knowledge have revitalized academic interest frameworks and highlighted the need for more understanding
in AI across various industries. For example, within a year of of AI acceptance among patients, health workers, and
its launch, OpenAI's Chat Generative Pre-Trained policymakers. Secinaro et al. (18) reviewed ethical challenges
Transformer (ChatGPT) gathered over 1000 citations in the related to AI in healthcare, such as accountability, privacy,
medical literature (7), showcasing the active interplay and transparency. The AI integration in healthcare presents
between AI innovation and academic research. Healthcare opportunities and challenges for industry and academia. (10)
leaders are using emerging technologies like the Internet of Our study aims to understand the barriers inhibiting the

IJISRT24OCT089 www.ijisrt.com 500

Volume 9, Issue 10, October– 2024 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165 https://doi.org/10.38124/ijisrt/IJISRT24OCT089

potential of AI implementation in healthcare. Extensive A. Aspects of AI model Design for Bridging the Gap between
literature exists concerning the challenges and support for Research and Implementation as Identified by Literature
implementing AI in healthcare. (19–21) However, a Review
significant portion of this body of knowledge is based on The long-term stability of any AI model performance
anecdotal evidence, narrative commentaries, small-scale raises concerns within health systems because of the absence
studies, and reviews lacking empirical support. of comprehensive guidance for implementing & post-
Consequently, our understanding of the contributing factors implementation maintenance procedures. (35) When we
of AI implementation success in healthcare remains observe the promising performance of an AI model during the
incomplete. (22) development phase, it raises the question of why there is a
disparity between this stage and the deployment phase. To
In the subsequent sections of our paper, we will explore resolve this implementation gap, we should redirect our focus
the fundamental aspects of the gaps and challenges associated from merely optimizing this area to encompass three crucial
with the implementation of AI in healthcare, using real-world & practical aspects of model design: actionability, safety, and
experiences. utility. (15) (Figure 1)

II. RESEARCH METHODOLOGY  Actionability

Clinicians or patients should be able to directly use the
This review was performed using specific keywords to AI algorithm results to take specific actions. Machine
search through scientific databases such as PubMed, Science learning models, though powerful, often lack clear follow-up
Direct, Google Scholar, and Springer Link. A thorough actions, leading to uncertainty about the next step. De Fauw
examination of relevant studies was conducted. The et al. conducted a deep learning optical coherence
keywords (“Artificial Intelligence,” “Implementations,” tomography scan interpretation study, in which they
“Applications,” “Barriers”, “AI in healthcare”, etc.) were segmented the scan and categorized several disorders, then
employed in various combinations to discover relevant gave the clinician a straightforward recommendation. (36) To
material. The titles and abstracts were evaluated to verify that ensure user-friendliness, we must connect machine learning
the papers focused specifically on the “Research gaps in AI,” tools to clinical actions and easily implement them into the
“Implementations of AI in healthcare,” and, “Challenges in clinical environment. (37)
implementations of AI in healthcare” The review
encompasses the data extracted from research papers,  Safety
reviews, systematic reviews, debates, workshops, and Patient safety should be a fundamental concern in the
conference papers published exclusively in the English design of medical AI models. The medical community is
language. familiar with the rigorous regulatory process for testing new
pharmaceutical or medical devices, but the safety of AI
III. RESULTS AND DISCUSSION algorithms is a significant concern because of issues such as
interpretability and external validity. (38) There is a lack of
Implementing AI-enabled tools in clinical settings has empirical evidence to demonstrate the safety and efficacy of
emphasized the need for ongoing monitoring and refinement algorithms in real-world settings. For widespread use, we
of prediction models. (23–26) It is increasingly evident that need empirical validation for ongoing algorithmic and
there is a temporal decline in model accuracy across diverse technical resilience. Model developers should engage with
clinical domains, despite the use of advanced AI algorithms. regulatory bodies and consider additional dimensions of
(27–30) The decrease in performance emphasizes the patient safety, such as protecting against algorithmic bias and
importance of ongoing improvement of AI-enabled tools, and model brittleness. It is vital to involve clinicians and patients
their clinical efficacy in upholding user trust and safety in in the process to ensure the translation of algorithms into true
decision-making processes. (23,27,29,31,32) The clinical benefit. (39)
performance of temporal models in real-time clinical settings
has been largely unexplored, with most studies focusing on  Utility
retrospective research data. (33) However, transitioning from The final step of any AI project should involve a cost-
research to real-time implementation can significantly impact utility assessment. This assessment compares the impact of
the performance, as inputs and data availability changes. (34). working with and without the algorithm, considering the
clinical and financial outcomes of false positives and
negatives. For instance, if we use an AI algorithm to screen
Electronic Health Records (EHR) for undiagnosed cases of a
rare disease like familial hypercholesterolemia, we need to
weigh the cost and utility of early detection against the
expenses and potential harm of false-positive cases and the
costs associated with deploying & maintaining the algorithm.
This model implementation necessitates an early and ongoing
execution of this utility assessment. (40)

IJISRT24OCT089 www.ijisrt.com 501

Volume 9, Issue 10, October– 2024 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165 https://doi.org/10.38124/ijisrt/IJISRT24OCT089

Fig 1 Summary of Three Aspects of AI model Design for Bridging the Gap between Research and Implementation as Identified
by Seneviratne et al. Paper (15)

B. The Gaps between Research and Implementation as burdensome financial obligations. The costs of AI integration
Identified by Literature Review are significant and may include talent acquisition, technology
In 2019, “Heart Flow” received FDA approval for its infrastructure upgrades, software procurement, and training
non-invasive, real-time, virtual modelling tool for coronary program implementation. This presents a substantial barrier
artery disease intervention. (41) However, FDA regulatory to the incorporate AI in the healthcare sector. (10)
processes can impede the development of such AI systems.
Comprehensive standards for approval processes will assist The development of laws and protocols helps AI
researchers and developers in meeting clearance requirements researchers to effectively present their work. TRIPOD-AI and
more quickly, decreasing delays and rejections of their PROBAST-AI, are a couple of examples which provide
applications. (42) The reporting standard presents a barrier to guidelines and tools to minimize research waste. (48) It is
implementing AI research in practical settings. Many reviews important to design AI applications with clinical needs in
have noted that studies creating prediction models for clinical mind to prevent research waste and align with clinicians'
use often lack transparent descriptions. This lack of requirements, even when predicting multiple outcomes
transparency can lead to a lack of trust among patients and simultaneously. (43)
clinicians and may limit the use and replication of these
findings. (43) In a Japanese study conducted by Changhee Han (49) et
al., their research team showed the clinically-relevant
Healthcare organizations that lack AI-compatible findings from their workshop by addressing three gaps
technologies might bridge the technology gap by upgrading between AI and Healthcare Sides. Their responses are
their existing systems to meet the integration criteria. The presented below:
lack of technical knowledge hold back the use of AI in
healthcare due to the gap among health professionals and the  GAP 1: AI, including Deep Learning, provides unclear
absence of requisite training. (44) It is important to decision criteria, does it make physicians reluctant to use
understand the precise function of AI to reduce health it for diagnosis in a clinical environment?
professionals' reluctance to adapt. (45) Data management and
security are critical components of AI integration, besides  Response on Healthcare side: "AI can aid in diagnosis but
technological and human challenges. (46) Enabling AI in should not replace physicians. It can provide a second
healthcare necessitates the availability of substantial amounts opinion by analyzing clinical data. However, its minimal
of medical data. Nevertheless, healthcare administration has explanation to persuade physicians and patients makes it
the potential to address the privacy and security of medical difficult. Physician intervention is necessary for an
data . (47) It is difficult for decision-makers to ensure the intuitive explanation. Pursuing explainable AI may
utmost level of protection while sharing medical data with AI decrease diagnostic accuracy, so physicians should still be
developers and tech experts. The economic implications are involved in making the final decision. Autonomous AI-
another obstacle that health organizations face. Particularly in based diagnosis without a physician is not the norm"
the public sector, healthcare organizations frequently endure

IJISRT24OCT089 www.ijisrt.com 502

Volume 9, Issue 10, October– 2024 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165 https://doi.org/10.38124/ijisrt/IJISRT24OCT089

 Response on AI side: "Deep Learning's explanations are  GAP 3: Is medical AI’s Diagnostic Accuracy Reliable?
not particularly poor compared to other systems or
physicians. We might be setting excessively high  Response on Healthcare side: "Assessing AI's diagnostic
standards for AI, promoting unnecessary anxiety. performance involves evaluating sensitivity, specificity,
Verifying the reliability of AI's diagnoses against and inter-scanner variability, as well as assessing its
physicians could make an intuitive explanation optional" suitability for clinical use"
 Response on AI side: "Research on medical AI is actively
 GAP 2 : Are there any Benefits to Actually Introducing focusing on robust datasets to reduce the risk of
Medical AI? overlooking diagnoses, prioritizing sensitivity over
specificity unless imbalance is disrupted"
 Response on Healthcare side: "AI's high accuracy and
convenience are commercially beneficial, especially in C. Barriers for Implementing AI in Healthcare
small clinics lacking CT or MR scanners, where Additionally, the objective of this paper is also to
physicians are desperately needed." identify and address the barriers that inhibit the healthcare
 Response on AI side: "Medical AI's commercial sector from implementing artificial intelligence (AI). The
deployment is linked to diagnostic accuracy, making it expanding body of literature underscores the significance of
profitable in upcoming years if it achieves outstanding investigating these challenges and offering scientifically
accuracy" based recommendations to facilitate the decision-making
process. We reviewed many of the included studies focussing
on various challenges associated with the deployment of AI
in healthcare. Implementing AI effectively presents
significant challenges, including ethical, technological,
safety, regulatory, workforce, social, and economic barriers.
(2,10) (Figure 2)

Fig 2 Common Barriers for Implementing AI in Healthcare

 Ethical Barrier: storage, and the security measures in place to protect their
Confidentiality, trust, consent, and conflicts of interest privacy. (51,52) He et al. recommended redefining patient
fall into the sub-units of ethical barrier. Developers need confidentiality and data privacy terms as data sharing
enormous datasets for training to create accurate AI systems, increases. (53) Healthcare AI implementation needs explicit
and there is a reasonable concern that this may violate patient governance that defines all stakeholders' data ownership. (44)
confidentiality. (50) AI developers must inform patients Marcu et al. emphasized the necessity of data privacy
about the fate of their data and obtain their consent for its use. regulations for secure data storage, usage, and sharing to
Patients need to understand who can access their data, the protect confidential information. (54)

IJISRT24OCT089 www.ijisrt.com 503

Volume 9, Issue 10, October– 2024 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165 https://doi.org/10.38124/ijisrt/IJISRT24OCT089

 Technological Barriers:  Social Barrier:

Integrating data in terms of accuracy, quality, and Healthcare bias, which involves treating patients
storage poses a big challenge. It is difficult to acquire and differently based on race, age, gender, or other factors, has
integrate large, diversified, high-quality data sets. (55). the potential to exacerbate this gap. Therefore, any new
Health organizations can address some of such challenges technology must not contribute to social inequality but
internally, giving them control over the causes. For instance, instead minimize healthcare system bias. Inequalities and
IT infrastructure constraints, which are significant in biases in healthcare can limit AI implementation. Studies
developing nations, influence AI integration. (17) Upgrades showed that social acceptance of AI models hinders
to meet integration requirements can help healthcare deployment. Singh et al. argued that AI implementation
businesses without AI-compatible systems to close this identifies important performance indices and tracks return on
technology gap. (10) investment to make healthcare more economical and
efficient. (47) However, the ‘black box’ nature of AI limits
 Safety Barrier: its social acceptance, and it is unclear whether patients will
We must carefully evaluate concerns about the impact accept a diagnosis from a computer instead of a clinician,
on patient safety throughout clinical AI implementation. (56) especially if it is perceived to save time and money without
Failure to address these safety concerns might severely clear evidence that quality has not been compromised. (2)
impact AI implementation, causing patients and staff to lose
trust. Patient safety must remain at the heart of AI for patient  Economic Barrier:
management and if any safety issues arise, especially for “Minimal funds. (Morris et al., 2023; Owoyemi et al.,
high-profile candidates, would immediately affect confidence 2020), Unclear Return and Investment in funds (ROI). (Wolff
in the technology and could inhibit its success. (2) et al., 2020), Development and maintenance charges.
(Alnasser, 2023), and service-based reimbursements. (Alami
 Regulatory Barrier: et al., 2020)” can be counted as some of the economic barriers
Whether regulations are in place or not, they can affect for implementing AI in healthcare. (10)
the implementation of AI in healthcare by both users and
developers. Systems regulators must understand and trust the The success, elevated accuracy, and promise that AI is
AIs used in their field to regulate effectively. Unfortunately, yielding in research are undeniable, despite its barriers. For
most current technologies lack transparency. (57) This major this advancement to be most beneficial in the field of
obstacle affects whether consumers and regulatory bodies are healthcare, it is essential to dismantle the barriers between AI
ready to trust an AI in clinical practice. The GDPR guidelines, research and clinical care. We can achieve this by
which became effective in May 2018, define a comprehensive implementing the discussed solutions and developing new
set of regulations for the governance of personal information. ones. (43)
These regulations have a wide range of implications for AI
implementation. (58) IV. CONCLUSION

 Workforce Barriers: The evolving theories, models, and frameworks in

The healthcare workforce's education and training implementation science hold significant promise in
significantly impede AI adoption. The medical community's advancing our grasp of AI implementation. This integrated
understanding of the potential and functionality of AI is still approach, which harmonizes AI and implementation science,
rudimentary, necessitating a significant measure for AI transcends the conventional limitations of each field. The
education and training. (59) We must address the challenges fusion of these distinct yet complementary disciplines is
of differing levels of technical literacy, and comprehension indispensable for gaining valuable insights into the bridging
of existing technologies before the workforce can become the gap for implementation of AI in healthcare. This paper,
settled and competent in clinical practice. (60) However, the while focusing on a general literature review, contributes to
difficulties could intensify as physicians strive to integrate AI the literature on bridging the gap between research and
platforms while battling with current healthcare technologies. implementation of AI in healthcare. However, further
(47) Healthcare providers should specifically design training broadened research in this area could explore more extensive
programs for physicians using AI systems and tailor these implementation of AI in healthcare.
programs to address various concerns that arise when using
unfamiliar technology. (61)

ABBREVIATIONS

AI Artificial Intelligence
ML Machine Learning
CHADS-VASC Congestive heart failure, Hypertension, Age, Diabetes mellitus, prior Stroke or TIA or
thromboembolism - Vascular disease, Age, Sex category
ChatGPT Chat Generative Pre-Trained Transformer
DL Deep Learning
FDA Food and Drug Administration

IJISRT24OCT089 www.ijisrt.com 504

Volume 9, Issue 10, October– 2024 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165 https://doi.org/10.38124/ijisrt/IJISRT24OCT089

GDPR General Data Protection regulation

IoT Internet of Things
IT Information Technology
NLP Natural Language Processing
CNN Convolutional Neural Networks
DCNN Deep Convolutional Neural Networks
DNN Deep Neural Networks
EHR Electronic Health Records
BEHRT Bidirectional Health Encoder Representations Transformer
TRIPOD-AI Transparent Reporting of a multivariable prediction model for Individual Prognosis or Diagnosis-
Artificial Intelligence
PROBAST-AI Prediction model Risk Of Bias ASsessment Tool – Artificial Intelligence

REFERENCES [10]. Aldwean A, Tenney D. Artificial Intelligence in

Healthcare Sector: A Literature Review of the
[1]. Prakash P. Artificial intelligence in healthcare: Adoption Challenges. OJBM. 2024;12(01):129–47.
Bridging the gap between boon and bane. IP Annals [11]. Trinkley KE, An R, Maw AM, Glasgow RE,
of Prosthodontics and Restorative Dentistry. Brownson RC. Leveraging artificial intelligence to
2024;2(10):90–2. advance implementation science: potential
[2]. Ahmed MI, Spooner B, Isherwood J, Lane M, Orrock opportunities and cautions. Implementation Sci. 2024
E, Dennison A. A Systematic Review of the Barriers Feb 21;19(1):17.
to the Implementation of Artificial Intelligence in [12]. Stone P, Brooks R, Brynjolfsson E, Calo R, Etzioni
Healthcare. Cureus [Internet]. 2023 Oct 4 [cited 2024 O, Hager G, et al. Artificial Intelligence and Life in
Jul 16]; Available from: https://www.cureus.com/ 2030: The One Hundred Year Study on Artificial
articles/170025-a-systematic-review-of-the-barriers- Intelligence [Internet]. arXiv; 2022 [cited 2024 Jul
to-the-implementation-of-artificial-intelligence-in- 16]. Available from: https://arxiv.org/abs/2211.
healthcare 06318
[3]. Ismail L, Materwala H, Karduck AP, Adem A. [13]. MacKay C, Klement W, Vanberkel P, Lamond N,
Requirements of Health Data Management Systems Urquhart R, Rigby M. A framework for
for Biomedical Care and Research: Scoping Review. implementing machine learning in healthcare based
J Med Internet Res. 2020 Jul 7;22(7):e17508. on the concepts of preconditions and postconditions.
[4]. Ismail L, Materwala H, Tayefi M, Ngo P, Karduck Healthcare Analytics. 2023 Nov;3:100155.
AP. Type 2 Diabetes with Artificial Intelligence [14]. Yu KH, Beam AL, Kohane IS. Artificial intelligence
Machine Learning: Methods and Evaluation. Arch in healthcare. Nat Biomed Eng. 2018 Oct
Computat Methods Eng. 2022 Jan;29(1):313–33. 10;2(10):719–31.
[5]. Chomutare T, Tejedor M, Svenning TO, Marco-Ruiz [15]. Seneviratne MG, Shah NH, Chu L. Bridging the
L, Tayefi M, Lind K, et al. Artificial Intelligence implementation gap of machine learning in
Implementation in Healthcare: A Theory-Based healthcare. BMJ Innov. 2020 Apr;6(2):45–7.
Scoping Review of Barriers and Facilitators. [16]. Ali O, Abdelbaki W, Shrestha A, Elbasi E, Alryalat
IJERPH. 2022 Dec 6;19(23):16359. MAA, Dwivedi YK. A systematic literature review of
[6]. McCarthy J, Minsky ML, Rochester N, Shannon CE. artificial intelligence in the healthcare sector:
A Proposal for the Dartmouth Summer Research Benefits, challenges, methodologies, and
Project on Artificial Intelligence, August 31, 1955. functionalities. Journal of Innovation & Knowledge.
AI Magazine. 2006 Dec 15;27(4):12–12. 2023 Jan;8(1):100333.
[7]. Temsah MH, Altamimi I, Jamal A, Alhasan K, Al- [17]. Petersson L, Larsson I, Nygren JM, Nilsen P, Neher
Eyadhy A. ChatGPT Surpasses 1000 Publications on M, Reed JE, et al. Challenges to implementing
PubMed: Envisioning the Road Ahead. Cureus artificial intelligence in healthcare: a qualitative
[Internet]. 2023 Sep 6 [cited 2024 Jul 16]; Available interview study with healthcare leaders in Sweden.
from: https://www.cureus.com/articles/186004- BMC Health Serv Res. 2022 Dec;22(1):850.
chatgpt-surpasses-1000-publications-on-pubmed- [18]. Secinaro S, Calandra D, Secinaro A, Muthurangu V,
envisioning-the-road-ahead Biancone P. The role of artificial intelligence in
[8]. Amjad A, Kordel P, Fernandes G. A Review on healthcare: a structured literature review. BMC Med
Innovation in Healthcare Sector (Telehealth) through Inform Decis Mak. 2021 Dec;21(1):125.
Artificial Intelligence. Sustainability. 2023 Apr [19]. Victor Mugabe K. Barriers and facilitators to the
14;15(8):6655. adoption of artificial intelligence in radiation
[9]. Pacis DMM, Subido EDC, Bugtai NT. Trends in oncology: A New Zealand study. Tech Innov Patient
telemedicine utilizing artificial intelligence. In Bali, Support Radiat Oncol. 2021 Jun;18:16–21.
Indonesia; 2018 [cited 2024 Jul 16]. p. 040009.
Available from: https://pubs.aip.org/aip/acp/
article/791769

IJISRT24OCT089 www.ijisrt.com 505

Volume 9, Issue 10, October– 2024 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165 https://doi.org/10.38124/ijisrt/IJISRT24OCT089

[20]. Strohm L, Hehakaya C, Ranschaert ER, Boon WPC, [33]. Davis SE, Lasko TA, Chen G, Matheny ME.
Moors EHM. Implementation of artificial Calibration Drift Among Regression and Machine
intelligence (AI) applications in radiology: hindering Learning Models for Hospital Mortality. AMIA Annu
and facilitating factors. Eur Radiol. 2020 Symp Proc. 2017;2017:625–34.
Oct;30(10):5525–32. [34]. Ötleş E. [PDF] Mind the Performance Gap:
[21]. Morrison K. Artificial intelligence and the NHS: a Examining Dataset Shift During Prospective
qualitative exploration of the factors influencing Validation | Semantic Scholar [Internet]. 2021 [cited
adoption. Future Healthc J. 2021 Nov;8(3):e648–54. 2024 Jul 16]. Available from:
[22]. Nilsen P. Making sense of implementation theories, https://www.semanticscholar.org/paper/Mind-the-
models and frameworks. Implement Sci. 2015 Apr Performance-Gap%3A-Examining-Dataset-Shift-
21;10:53. %C3%96tle%C5%9F-
[23]. Amarasingham R, Patzer RE, Huesch M, Nguyen Oh/bd6d24be7320d7d456e610827b89639f3249a6ab
NQ, Xie B. Implementing Electronic Health Care [35]. Davis SE, Walsh CG, Matheny ME. Open questions
Predictive Analytics: Considerations And and research gaps for monitoring and updating AI-
Challenges. Health Affairs. 2014 Jul;33(7):1148–54. enabled tools in clinical settings. Front Digit Health.
[24]. Smith J. Setting the agenda: an informatics-led policy 2022 Sep 2;4:958284.
framework for adaptive CDS. Journal of the [36]. De Fauw J, Ledsam JR, Romera-Paredes B, Nikolov
American Medical Informatics Association. 2020 S, Tomasev N, Blackwell S, et al. Clinically
Dec 9;27(12):1831–3. applicable deep learning for diagnosis and referral in
[25]. Jenkins DA, Martin GP, Sperrin M, Riley RD, retinal disease. Nat Med. 2018 Sep;24(9):1342–50.
Debray TPA, Collins GS, et al. Continual updating [37]. Sittig DF, Singh H. A new sociotechnical model for
and monitoring of clinical prediction models: time for studying health information technology in complex
dynamic prediction systems? Diagn Progn Res. 2021 adaptive healthcare systems. Qual Saf Health Care.
Dec;5(1):1. 2010 Oct;19 Suppl 3(Suppl 3):i68-74.
[26]. Petersen C, Smith J, Freimuth RR, Goodman KW, [38]. Knight W. The Dark Secret at the Heart of AI
Jackson GP, Kannry J, et al. Recommendations for [Internet]. 2017. Available from:
the safe, effective use of adaptive CDS in the US https://www.technologyreview.com/2017/04/11/511
healthcare system: an AMIA position paper. Journal 3/the-dark-secret-at-the-heart-of-ai/
of the American Medical Informatics Association. [39]. Keane PA, Topol EJ. With an eye to AI and
2021 Mar 18;28(4):677–84. autonomous diagnosis. npj Digital Med. 2018 Aug
[27]. Finlayson SG, Subbaswamy A, Singh K, Bowers J, 28;1(1):40, s41746-018-0048-y.
Kupke A, Zittrain J, et al. The Clinician and Dataset [40]. Banda JM, Sarraju A, Abbasi F, Parizo J, Pariani M,
Shift in Artificial Intelligence. N Engl J Med. 2021 Ison H, et al. Finding missed cases of familial
Jul 15;385(3):283–6. hypercholesterolemia in health systems using
[28]. Hickey GL, Grant SW, Murphy GJ, Bhabra M, machine learning. npj Digit Med. 2019 Apr
Pagano D, McAllister K, et al. Dynamic trends in 11;2(1):23.
cardiac surgery: why the logistic EuroSCORE is no [41]. Qayyum S, Habib A, Kechyn S, Auger D, Damani N,
longer suitable for contemporary cardiac surgery and Griguer A, et al. P01 Heartflow: experience of a high-
implications for future risk models. European Journal volume district general hospital. In: Scientific poster
of Cardio-Thoracic Surgery. 2013 Jun 1;43(6):1146– abstracts [Internet]. BMJ Publishing Group Ltd and
52. British Cardiovascular Society; 2020 [cited 2024 Jul
[29]. Davis SE, Lasko TA, Chen G, Siew ED, Matheny 16]. p. A5.3-A5. Available from:
ME. Calibration drift in regression and machine https://heart.bmj.com/lookup/doi/10.1136/heartjnl-
learning models for acute kidney injury. Journal of 2020-BSCI.14
the American Medical Informatics Association. 2017 [42]. Artificial Intelligence and Machine Learning in
Nov 1;24(6):1052–61. Software as a Medical Device | FDA [Internet]. [cited
[30]. Wong A, Cao J, Lyons PG, Dutta S, Major VJ, Ötleş 2024 Jul 16]. Available from:
E, et al. Quantification of Sepsis Model Alerts in 24 https://www.fda.gov/medical-devices/software-
US Hospitals Before and During the COVID-19 medical-device-samd/artificial-intelligence-and-
Pandemic. JAMA Netw Open. 2021 Nov machine-learning-software-medical-device
19;4(11):e2135286. [43]. Shipley E, Joddrell M, Lip GY, Zheng Y. Bridging
[31]. Jiang X, Osl M, Kim J, Ohno-Machado L. Calibrating the Gap Between Artificial Intelligence Research and
predictive model estimates to support personalized Clinical Practice in Cardiovascular Science: What the
medicine. J Am Med Inform Assoc. 2012 Clinician Needs to Know. Arrhythm Electrophysiol
Mar;19(2):263–74. Rev [Internet]. 2022 Apr 14 [cited 2024 Jul 16];11.
[32]. Van Calster B, Vickers AJ. Calibration of Risk Available from: https://www.aerjournal.com/
Prediction Models: Impact on Decision-Analytic articleindex/aer.2022.07
Performance. Med Decis Making. 2015 [44]. Reddy S, Allan S, Coghlan S, Cooper P. A
Feb;35(2):162–9. governance model for the application of AI in health
care. Journal of the American Medical Informatics
Association. 2020 Mar 1;27(3):491–7.

IJISRT24OCT089 www.ijisrt.com 506

Volume 9, Issue 10, October– 2024 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165 https://doi.org/10.38124/ijisrt/IJISRT24OCT089

[45]. Weinert L, Müller J, Svensson L, Heinze O. [56]. Challen R, Denny J, Pitt M, Gompels L, Edwards T,
Perspective of Information Technology Decision Tsaneva-Atanasova K. Artificial intelligence, bias
Makers on Factors Influencing Adoption and and clinical safety. BMJ Qual Saf. 2019
Implementation of Artificial Intelligence Mar;28(3):231–7.
Technologies in 40 German Hospitals: Descriptive [57]. Laï MC, Brian M, Mamzer MF. Perceptions of
Analysis. JMIR Med Inform. 2022 Jun artificial intelligence in healthcare: findings from a
15;10(6):e34678. qualitative survey study among actors in France. J
[46]. Firouzi F, Farahani B, Barzegari M, Daneshmand M. Transl Med. 2020 Dec;18(1):14.
AI-Driven Data Monetization: The Other Face of [58]. UK GDPR guidance and resources | ICO [Internet].
Data in IoT-Based Smart and Connected Health. [cited 2024 Jul 16]. Available from:
IEEE Internet Things J. 2022 Apr 15;9(8):5581–99. https://ico.org.uk/for-organisations/uk-gdpr-
[47]. Singh RP, Hom GL, Abramoff MD, Campbell JP, guidance-and-resources/
Chiang MF, on behalf of the AAO Task Force on [59]. Paranjape K, Schinkel M, Hammer RD, Schouten B,
Artificial Intelligence. Current Challenges and Nannan Panday RS, Elbers PWG, et al. The Value of
Barriers to Real-World Artificial Intelligence Artificial Intelligence in Laboratory Medicine.
Adoption for the Healthcare System, Provider, and American Journal of Clinical Pathology. 2021 May
the Patient. Trans Vis Sci Tech. 2020 Aug 11;9(2):45. 18;155(6):823–31.
[48]. Collins GS, Dhiman P, Andaur Navarro CL, Ma J, [60]. Panch T, Mattie H, Celi LA. The “inconvenient truth”
Hooft L, Reitsma JB, et al. Protocol for development about AI in healthcare. npj Digit Med. 2019 Aug
of a reporting guideline (TRIPOD-AI) and risk of 16;2(1):77.
bias tool (PROBAST-AI) for diagnostic and [61]. Scarpazza C, Ha M, Baecker L, Garcia-Dias R,
prognostic prediction model studies based on Pinaya WHL, Vieira S, et al. Translating research
artificial intelligence. BMJ Open. 2021 findings into clinical practice: a systematic and
Jul;11(7):e048008. critical review of neuroimaging-based clinical tools
[49]. Han C, Rundo L, Murao K, Nemoto T, Nakayama H. for brain disorders. Transl Psychiatry. 2020 Apr
Bridging the Gap Between AI and Healthcare Sides: 20;10(1):107.
Towards Developing Clinically Relevant AI-
Powered Diagnosis Systems. In: Maglogiannis I,
Iliadis L, Pimenidis E, editors. Artificial Intelligence
Applications and Innovations [Internet]. Cham:
Springer International Publishing; 2020 [cited 2024
Jul 16]. p. 320–33. (IFIP Advances in Information
and Communication Technology; vol. 584).
Available from: https://link.springer.com/
10.1007/978-3-030-49186-4_27
[50]. Grant K, McParland A, Mehta S, Ackery AD.
Artificial Intelligence in Emergency Medicine:
Surmountable Barriers With Revolutionary Potential.
Annals of Emergency Medicine. 2020
Jun;75(6):721–6.
[51]. Brady AP, Neri E. Artificial Intelligence in
Radiology—Ethical Considerations. Diagnostics.
2020 Apr 17;10(4):231.
[52]. Kelly CJ, Karthikesalingam A, Suleyman M, Corrado
G, King D. Key challenges for delivering clinical
impact with artificial intelligence. BMC Med. 2019
Dec;17(1):195.
[53]. He K, Gan C, Li Z, Rekik I, Yin Z, Ji W, et al.
Transformers in medical image analysis. Intelligent
Medicine. 2023 Feb;3(1):59–78.
[54]. Marcu LG, Boyd C, Bezak E. Current issues
regarding artificial intelligence in cancer and health
care. Implications for medical physicists and
biomedical engineers. Health Technol. 2019
Aug;9(4):375–81.
[55]. Ben-Israel D, Jacobs WB, Casha S, Lang S, Ryu
WHA, De Lotbiniere-Bassett M, et al. The impact of
machine learning on patient care: A systematic
review. Artificial Intelligence in Medicine. 2020
Mar;103:101785.

IJISRT24OCT089 www.ijisrt.com 507