Proceedings of the International Conference on Self Sustainable Artificial Intelligence Systems (ICSSAS 2023)

IEEE Xplore Part Number: CFP22DN7-ART; ISBN: 979-8-3503-0085-7

Comparative Analysis of Machine Learning Models

for Early Prediction of Cancer
1 2
A.Elangovan, Dr.M.Nagarajan,
Research Scholar, Dept.of ECE, Associate Prof, Dept.of ECE,
Bharath Institute of Higher Education and Research, Chennai. Bharath Institute of Higher Education and Research, Chennai
[email protected] [email protected]
2023 International Conference on Self Sustainable Artificial Intelligence Systems (ICSSAS) | 979-8-3503-0085-7/23/$31.00 ©2023 IEEE | DOI: 10.1109/ICSSAS57918.2023.10331672

Abstract— A deadly and occasionally fatal condition, cancer drawing inferences [9]. These methods can be modified for
may have a number of detrimental effects on the body, rational decision-making in the medical field.
represents one among the main causes of death, and is getting Early cancer research has many requirements that
harder and harder to diagnose. Every type of cancer involves a must be met in order to improve patient medical treatments
unique group of characteristics, signs, and treatments, and early [10]. In order to categorize cancer patients as low or
detection and care are crucial for a good prediction. On the basis
of the type as well as position of the tumor, doctors employ a
high risk, numerous studies investigated the use of ML as
range of methods to diagnose cancer. One among the methods well as DL methodologies in bioinformatics and
doctors utilize to identify cancer is imaging testing, like MRI biochemistry [11]. The need to categorize cancer patients into
scans, CT scans, X-rays, and PET scans, which may offer exact low- or high-risk categories has prompted several research
images of the body's internal components to notice any teams from the bioinformatics and biomedical sectors to
discrepancies. By emphasizing the importance of ML and investigate the application of ML and DL approaches [12]. In
DL methods like KNN, SVM, NB, DT, Deep Boltzmann a bid to imitate the onset and treatment of malignant illnesses,
machine, DNN, and others, this survey determines several techniques have been used. Additionally, the ability
computational-intelligence techniques and offers a way to of ML and DL algorithms to identify critical characteristics
influence future work.
in challenging datasets illustrates their importance [13]. The
Keywords— Cancer Prediction, Early stage, Machine Learning,
application of ML for cancer prognosis as well as forecast is
Deep Learning. relatively modern. Since it conforms to an increasing trend
towards personalized, predictive therapy, the latter technique
I. INTRODUCTION is particularly appealing [14]. The development of forecast
methods that forecast the arrival of a cancer cure frequently
Globally, cancer is increasingly one among the employs several of these techniques [15]. The early cancer
leading reasons of mortality [1]. According to estimates, prediction model can be analyzed using the steps as shown in
various cancers claimed the lives of almost 9.6 million the figure 1below.
individuals in 2019. Every year, thousands of people suffer
and die owing to errors in the healthcare frameworks across
the globe [2]. The genes that control cell proliferation as well
as the anomalies that arise throughout the enhancement of a
particular cancer are detailed in a genetic profile [3].
Extraction of this important data may help with accurate
illness onset forecast and management strategies for drug
selection as well as individualized care [4]. The accuracy of
predictions is crucial for making wise decisions;
hence science's major focus is on addressing the limitations
of human discoveries as well as assessments [5].
Many tools as well as methods known as DSS have
been created in the domains of information science, AI, etc.
for complicated decision making [6]. The concept of DSS is Fig 1 Step analysis for the early cancer prediction model
quite broad, and each researcher will characterize it
differently. DSSs are developing a solid reputation in several
fields, including health, business, and the military, among The contribution of the study is shown below.
others [7]. It is especially valued in situations when • To perform a survey by gathering the state-
optimality and accuracy are important, as well as when the of-the-works associated with the early
volume of information being offered is too much for the cancer prediction.
perception of an unconstrained human judgement [8]. DSS • To analyze various ML and DL methods
may support human rational reasoning by synthesizing used for early cancer prediction.
numerous knowledge sources, while providing rapid • To conclude the survey by providing a
accessibility to associated data and aiding the process of direction for future research.

979-8-3503-0085-7/23/$31.00 ©2023 IEEE 325

Authorized licensed use limited to: MICROSOFT. Downloaded on December 28,2024 at 07:34:27 UTC from IEEE Xplore. Restrictions apply.
 Proceedings of the International Conference on Self Sustainable Artificial Intelligence Systems (ICSSAS 2023)
IEEE Xplore Part Number: CFP22DN7-ART; ISBN: 979-8-3503-0085-7

The organization of the study is given as follows. cancer. A person with cancer may also experience emotional,
Section I shows the introduction of the early cancer financial, and physical effects on themselves as well as their
prediction model. Section II is the background of cancer. family. Nausea and fatigue are two physical side impacts of
Section III is the types of cancer. Section IV is the techniques cancer therapy that may be treated with medicine and
in ML. Section V is the techniques in DL. Section VI is the various therapies. For those coping with cancer, emotional
applications of smart phone in cancer diagnosis and therapy. assistance like therapy and support groups might be
Section VII is open challenges with distinct methods. Section beneficial. For those stricken by cancer, financial help in the
VIII is problem definition. Section IX is related works. form of financial aid programmes and insurance coverage
Section X is comparative analysis. Section X1 is summary. might be crucial.

II. BACKGROUND OF CANCER III. TYPES OF CANCER

The Latin term "carcinoma" (which refers to "a There are about 100 distinct types of cancer. Cancer
crab") is where the word "cancer" first appeared. This word may also be described in terms of the kind of cells that cause
was utilized since cancer cells may propagate and infiltrate it, like squamous or epithelial cells.
different areas of the body and frequently imitate the legs of Ovarian cancer: On the basis of the stage and type
a crab. There exist several distinct cancer forms, and every of the disease, the symptoms and signs of ovarian cancer
format is categorized according to the kind of cell that is may differ, still familiar ones involve a sudden feeling or
originally harmed. Leukemia, prostate cancer, bloating of fullness, pelvic or abdominal pain or
lymphoma, and colon cancer are a few further prevalent swelling, weight loss or loss of appetite, variations in bladder
cancers. or bowel habits, back pain, exhaustion or weakness, and
Although some genetic alterations may make an menstrual variations.
individual more likely to have cancer, cancer is not usually Ovarian cancer typically progresses through four
the result of these abnormalities. An individual's chance of stages: In stage 1, the sole sources of the cancer
acquiring cancer may also be increased by environmental represents one or both ovaries; Stage 2 cancer means that it
elements, like exposure to specific substances or chemicals. has spread to nearby tissues, like the fallopian tubes or uterus;
Cancer may also be brought on by certain viruses, like the Stage 3 cancer means that it has progressed to nearby lymph
HPV. Cancer growth is also influenced by variables related nodes, the abdominal lining; Stage 4 is when the cancer has
to culture. For instance, smoking increases the chance of spread to other parts.
developing numerous cancers, such as throat, bladder, and Prostate cancer: It usually grows slowly over years
lung cancer. Additionally, excessive alcohol use raises the in males and represents a common type of cancer.
chance of developing several cancers, including liver and Difficulty urinating, a weak urine stream, lower
breast cancer. back or pelvic discomfort, blood in the semen or
Periodic cancer screenings, like colonoscopies for urine, difficulty getting or maintaining an erection or erectile
colon cancer and mammograms for breast cancer, may also dysfunction, and fractures or bone pain are possible signs of
aid in the early (and major curable) detection of the prostate cancer.
disease. Normally, an assortment of biopsy, laboratory Brain cancer: On the basis of the stage and kind of
testing, and medical imaging is used to diagnose cancer. the disease, different brain cancers may present with different
During a biopsy, a specimen of tissue is removed from a symptoms and warning signs. Brain cancer is frequently
questionable location and analyzed beneath a microscope to classified into four classes on the basis of how the tumor
see whether it contains any malignant cells. The position as grows: First grade: The slow-growing cancer cells look
well as size of a tumor may be determined with the use of similar to healthy cells; Second grade: The cancer cells seem
medical imaging techniques including CT scans, X-rays, and minimal natural and grow much swiftly; In grade 3, cancer
MRIs. Blood as well as urine analyzes in the laboratory cells rapidly grow and vary noticeably from healthy cells in
may both be employed to identify cancer. terms of morphology; The cancer cells have a high level of
Finding the cancer's stage represents the next abnormalities in grade 4, are expanding swiftly, and are
phase once a diagnosis has been established. The size spreading.
associated with the tumor and whether it has spread to Pancreatic cancer: Cancer stages and symptoms
various bodily areas are considered the cancer's phase. For may vary on the basis of the kind as well as stage of the
choosing the most effective course of therapy, this data is disease. The common symptoms of pancreatic cancer include
crucial. Radiation therapy, chemotherapy, and surgery are loss of weight or appetite, stomach or back pain, exhaustion
frequently used in the therapy of cancer. Surgery may be or weakness, vomiting or nausea, light-colored or dark urine
utilized to treat various different forms of cancer by faeces, jaundice (yellowing of the eyes or skin), and newly
removing malignant tissue. The kind as well as phase diagnosed diabetes.
associated with the cancer, as well as the patient's general Pancreatic cancer is often categorized into four
condition, will completely influence the therapy option. stages: The pancreas represents the sole organ in which the
Targeted treatment or immunotherapy can also be disease has spread in stage 1; The pancreas is yet there at
employed to combat cancer in specific circumstances. A stage 2; In stage 3, the cancer has spread to the
form of therapy known as targeted therapy targets particular stomach, spleen, liver, or remaining nearby organs or blood
proteins, genes, or the conditions in the tissue that support the vessels; Stage 4 cancer refers to the spread of the disease to
development of cancer cells. Treatments like immunotherapy new organs or tissues, like the bones or lungs.
support the body's immune response in its battle against

979-8-3503-0085-7/23/$31.00 ©2023 IEEE 326

Authorized licensed use limited to: MICROSOFT. Downloaded on December 28,2024 at 07:34:27 UTC from IEEE Xplore. Restrictions apply.
 Proceedings of the International Conference on Self Sustainable Artificial Intelligence Systems (ICSSAS 2023)
IEEE Xplore Part Number: CFP22DN7-ART; ISBN: 979-8-3503-0085-7

Liver cancer: Liver cancer often occurs in the liver, patient's unique requirements and traits, this may be utilized
the biggest organ in the body and a vital organ for purifying to optimize cancer therapeutic programmes.
as well as digesting blood. Unsupervised learning: Unlabeled
Usually, there are several stages of liver cancer: information must be fed to a methodology in order for it to
Stage 1 refers to the early stages of cancer that have not identify patterns as well as correlations on its own. By doing
spread past the liver; Stage 2 cancer is yet restricted to the so, it can be possible to spot odd or unexpected trends in
liver while being more enhanced than stage 1 cancer; Stage 3 medical information that might point to the existence of
cancer has progressed to nearby blood vessels or organs like cancer.
the gallbladder or pancreatic; stage 4 cancer has migrated to Supervised learning: Building a methodology
remaining organs including the lungs, bones, or brain. employing labelled information, in which every input is
Lung cancer: Depending on the kind and degree of matched with the proper output, is required for this.
the disease, lung cancer stages and symptoms might vary, but In general, machine learning approaches may be
common signs include chest pain, persistent highly effective instruments for increasing the speed as well
cough, exhaustion, and wheezing. as precision of cancer detection.
It is possible to stage NSCLC in the below methods: LR: To predict the likelihood that a patient will
Stage 1 cancer is limited to the lung and has not progressed acquire a certain type of cancer, the statistical technique of
to the lymph nodes or any remaining organs; Stage 2 cancer LR is extensively employed in cancer diagnosis. The
means that it has spread to thoracic lymph nodes or adjacent advantage of using LR is that numerous factors may be
tissues; Stage 3 cancer refers to when the disease has analyzed simultaneously, making it easier to find the
advanced to organs or middle-chest lymph nodes; In stage 4, major reliable cancer predictors. The detection of cancer,
the cancer has spread to remaining organs, such as the bones which can be impacted by a number of risk factors,
or liver, among others. involving lifestyle, family history, age, and environmental
Oral cancer: In the tissues of the throat or mouth, variables, might greatly benefit from this.
oral cancer may be seen. The roof, tongue, cheeks, gums, Ensemble models: The forecasts of several separate
lips, or floor of the mouth could all be affected. individual methods, each of which may have a distinct area
Oral cancer progresses via three stages: Stage I of specialization, can be combined using ensemble methods.
cancers are mild; Stage II cancers frequently range 2-4 cm This may improve the overall accuracy as well as robustness
and are more enhanced than stage I cancers; Stage III cancers of the forecasts while lowering the risk of overfitting to the
have either migrated to nearby lymph nodes; In stage IV, the training information. When a high degree of precision is
cancer has spread to many organs. needed in cancer detection and the information are
Oral cancer symptoms include pain while chewing complicated or noisy, this may be especially helpful.
or swallowing, lumps or soreness in the mouth that does not SVM: SVM is capable of managing complicated
go away, numbness in the face, loose teeth, chronic foul data distributions, involving information that cannot be
breath, and difficulty maintaining dentures. separated linearly. This is crucial in the detection of cancer
Breast cancer: Breast cancer normally originates in since the link between the input data and the result might be
breast tissue in women and, less commonly, in males. Breast extremely complicated and difficult to record using more
cancer signs include redness, swelling, thickening or warmth basic methods.
in the breast, variations in the breast's shape or size, RF: Even if few among the input features are only
puckering or dimpling of the breast skin, variations to the weakly connected with the result, RF may accommodate
nipple. features. This is crucial in the detection of cancer since there
Breast cancer typically progresses through five can be several variables that affect a patient to assess or have
stages: Stage 0 cancers are non-invasive and only affect the a tenuous connection to the result.
lobules or milk ducts of the breast tissue; A cancer that is KNN: One possible benefit of KNN is that it may be
stage 1 is small; Stage 2 cancer is still contained inside the helpful for detecting uncommon cancer subtypes as well as
breast tissue; In stage 3, the cancer has spread to the lymph instances. This is due to categorization method, in which
nodes and additional chest wall or breast tissues; In stage 4, every novel example is categorized on the basis of its K
the cancer has spread to remaining organs. closest neighbors’ similarities. KNN offers a more precise
diagnosis if the patient's signs or test findings
IV. TECHNIQUES IN ML vary significantly from those of remaining patients in the
dataset.
Numerous machine learning methods may be DT: DTs may offer light on which elements are
utilized to the cancer diagnostic process. Few of these most crucial in reaching a detection and are relatively simple
methods consist of: to read. This is particularly beneficial for professionals who
Transfer learning: This approach employs a must decide how to continue with the treatment of a patient.
methodology that has already been trained as the basis for a NB: A probabilistic as well as classification method
novel job rather than creating a methodology from the start. called NB determines the likelihood that a patient has cancer
When there exists a dearth of training information, the pre- based on their signs or test findings. Utilizing NB for cancer
trained methodology might be a useful starting point for detection has a number of benefits, one among which is its
operations. ability to process vast volumes of information quickly and
Reinforcement learning: This comprises guiding a effectively. It works with a range of data types and is also
methodology's decision-making in a situation by rewarding rather easy to set up.
or penalizing particular behaviors. On the basis of each

979-8-3503-0085-7/23/$31.00 ©2023 IEEE 327

Authorized licensed use limited to: MICROSOFT. Downloaded on December 28,2024 at 07:34:27 UTC from IEEE Xplore. Restrictions apply.
 Proceedings of the International Conference on Self Sustainable Artificial Intelligence Systems (ICSSAS 2023)
IEEE Xplore Part Number: CFP22DN7-ART; ISBN: 979-8-3503-0085-7

ANN: Cancerous tumors may be found using which may enhance patient findings and lessen the
ANNs taught to analyze clinical images from CT scans, X- requirement for intrusive operations or therapies.
rays, and MRIs. ANNs may be trained to analyze genomic RNN: A recurrent connection is used by a RNN to
information and find mutations linked to cancer. handle sequential information. The RNN's intermediary step
Hybrid models: The detection as well describes the state vector. The state vector from the past is
as management of cancer are increasingly using hybrid ML evaluated along with the values from the current input. The
frameworks. Hybrid models, which integrate the benefits of input from both the present as well as the past influences the
several ML methods, may improve the accuracy of cancer result. In general, RNNs may be a helpful methodology for
detection and support the forecast of patient outcomes. This the detection of cancer, especially when it's necessary to
may help healthcare practitioners choose the best effective examine sequential information and identify intricate
treatment plan for patients after conducting additional patterns in the information. RNNs are simply one
research. among various instruments available for the detection of
cancer, and the precise technique or mixture of techniques
V. TECHNIQUES IN DL employed will vary based on the requirements as well
as circumstances of the situation.
The various techniques of deep learning used for Hybrid Models: Cancer detection as well
predicting cancer at an early stage is described as follows. as treatment may be difficult and drawn out, involving the
DBM: A deep learning methodology called a DBM knowledge of several different medical specialists as well
may be applied to quickly and effectively analyze massive as the usage of several diagnostic techniques. The accuracy
volumes of complicated information, like gene expression as well as effectiveness of cancer detection have shown
information. This can aid in locating signatures or patterns potential for hybrid deep learning methods. It includes a
that can be linked to certain cancer formats, leading to more variety of information kinds and origins, including EHRs,
precise as well as individualized diagnosis. medical images, and genetic information, represents one
DCNN: DCNNs represents a common type of benefit of employing them to diagnose cancer. These
ANNs in the domains of image analysis and computer vision. methods may give a more thorough image features by
This may enable early identification and much more continuously analyzing many forms of information, enabling
successful treatment by assisting physicians as well more focused as well as individualized treatment strategies.
as radiologists in identifying possible malignant regions that
could else go undetected. VI. RELATED WORKS
DBN: Researchers have created DBNs, a form of
ANNs, to analyze complex information and produce In 2021, Shakir et al. [16] have created a Bayesian
forecasts. With benefits as well as drawbacks, DBNs have inversion methodology for longitudinal chest CT images that
been used in a number of domains, involving the detection of may effectively classify lung cancer into several categories.
cancer. This may aid physicians in making better-informed The effectiveness of lung cancer classifiers is hindered by the
choices regarding patient treatment as well as enhancing lack of a significant count of training medical images. The
results. chance of cancer as well as its pathological phase may be
DNN: A sort of ML methodology called DNNs may calculated using a combined likelihood function that
be taught to examine complicated datasets, such as medical incorporates diagnostic radiomic characteristics.
and patient image information. DNNs have demonstrated In 2019, Waseem et al. [17] have optimized three
potential in the study of medical imagery like CT scans and parameters instead of focusing on the two conventional
mammograms, which are used in the detection of cancer. parameters (rejection rate and learning algorithm). To
DNNs may be effective in the identification of cancer examine the robust forecasting of cancer, various FS
because they may identify nuances in medical images that techniques—involving the Relief, and IG, LVF, t-test —and
human eyes overlook. This might improve the effectiveness RO classifiers are analyzed on various rejection criteria. Each
and accuracy of cancer diagnoses, which will eventually be of the three cancer datasets leukaemia, (breast, and colon)
advantageous to patients as well as medical professionals. was employed to analyze the prediction of cancer utilizing a
LSTM: One among the numerous applications for distinct RO classifier after being minimized employing a
LSTM RNNs represents the diagnosis of cancer. The distinct FS approach. The findings show that distinct FS
capacity of LSTM to analyze intricate time-series approaches have varying RO classifier prediction accuracies
information, like patient medical records or gene expression for every dataset.
patterns, describes one benefit of applying it in cancer In 2018, Raweh et al. [18] have used a hybridized
detection. Massive amounts of information can be processed strategy to attempt to forecast cancer. This technique also
using LSTM, which may also spot patterns that could point utilizes the high-dimensionality issue associated with DNA
to the existence of cancer or remaining diseases. This could methylation information. The findings indicate that
result in earlier detection and much more precise diagnosis, classification accuracy generally increases, indirectly
which might enhance patient findings. supporting dependability.
Deep Autoencoder: The deep autoencoder In 2022, Qi et al. [19] have included 3411
methodology represents a well-liked deep learning strategy to respondents. By including 28 variables into methods created
extract characteristics from the initial mass spectral using PCA and ANN technologies, cancer may be predicted.
information. The mass spectrum information is used as the While unsupervised training and extracting pertinent
input as well as output vectors after preprocessing. This may characteristics are done using DBN. Supervised classification
result in early cancer identification and a better diagnosis, was carried out and DBN was optimized using the ELM. The

979-8-3503-0085-7/23/$31.00 ©2023 IEEE 328

Authorized licensed use limited to: MICROSOFT. Downloaded on December 28,2024 at 07:34:27 UTC from IEEE Xplore. Restrictions apply.
 Proceedings of the International Conference on Self Sustainable Artificial Intelligence Systems (ICSSAS 2023)
IEEE Xplore Part Number: CFP22DN7-ART; ISBN: 979-8-3503-0085-7

supervised fine tuning was done via the BP technique. In order to mimic the activity of genes, every method focused
Furthermore, methods for predicting familiar tumors using on a specific group of parameters. The recommended
DBN-ELM-BP and PCA-ANN were developed. The findings method can be trained and predicted effectively with
demonstrate that the DBN-ELM-BP method could offer a minimal parameters.
non-invasive, cost-effective way for doctors to
perform diagnostic judgements that could forecast the risk of Table 1 Some related cancer prediction models
familiar malignancies.
In 2011, Jha et al. [20] created a Ref. No. Technique Pros Cons
combined continuous-flow microfluidic device on a glass
substrate having PCR, cell lysis, and micro channels made of Kourou et ML The It does not provide
al. [26]
PDMS. During the PCR procedure, a micro heater made of findings a reason for the
ITO was employed for thermal cycling while a gold- are disease findings.
microelectrode was utilized for electrochemical cell lysis. revealed in
Fewer than 45 minutes were needed to complete the full an accurate
response. manner.
In 2022, Ganesh et al. [21] have discovered an Hunter et AI It reveals It is restricted from
uncommon subgroup of tumor cells called MICs, which al. [27]
the data security,
could signal the beginning of cancer metastasis regardless of possible resource demands,
the kind of cancer, the tissue from which it originated, or the clinical ethical concerns,
phase of the cancer. MICs were cancer non-specific and not implicatio and reporting
dependent on epigenetic makeup. ns for the standards.
In 2022, McDonnell et al. [22] have AI
comprised individuals having a validated novel cancer. algorithms
Initial diagnostic follow-up was directed by a .
multidisciplinary review team. The workup necessary to get Shaikh ML and The It does not produce
at a conclusive detection was guided by clinical knowledge and Rao
DL forecasts relevant findings
from the CSO forecast. [28]
are and combined
In 2022, Matsuzaki et al. [23] have assessed the addressed clinical decision-
effectiveness of a blood miRNome-oriented classifier to on the making.
distinguish between different cancer kinds basis of
utilizing sophisticated ML. 5,013 non-cancer specimens as data
well as 7,931 blood specimens from individuals who had 13 samples
different forms of solid malignancies made up the training supervisio
group. 1,256 non-cancer specimens and 1,990 cancer n.
specimens made up the validation group. Every miRNA's Rehaman SMOTE It The main
influence on the categorization of cancer types was assessed, et al. [29]
minimizes drawback is its
and those having significant benefits were discovered. This the small sample size.
work suggests that the creation of a blood-oriented cancer imbalance
classification framework might be facilitated by large-scale associated
serum miRNomics in conjunction with ML. The with the
development of floating miRNA detection for usage in derivative
clinical settings may be facilitated by additional research into dataset.
the regulatory processes of the miRNAs that significantly Jain et al. ML and Computati It does not address
aided in the correct forecast of cancer kind. [30]
DL onal- various clinical
In 2023, Klein et al. [24] have incorporated patients intelligenc issues.
who had cancer signals found in any among the two iterations e
of an MCED screening that PATHFINDER examined. techniques
Predicted diagnostic assessments by cancer signal origin are
forecast were defined using predetermined criteria that were evaluated.
blind to research data. The agreement among cancer signal Premalath Novel The risk It consumes more
origin forecasts and the actual diagnostic assessment was a et al.
multi levels are time.
next assessed using these parameters. A year following [31]
layered analyzed
enrolment, the trial's EOS cancer condition was evaluated to
method to support
see if an absolute cancer detection had been made or not.
prognosis.
After a diagnostic workup focused on the cancer signal
origin, the majority of patients had their diagnosis resolved.
There was yet a risk of cancer in patients having previous
diagnoses of cancer despite negative or ambiguous first
assessments, indicating the requirement for further testing.
In 2023, Gangurde et al. [25] employed NPR and
CNN techniques to identify and classify the kind of cancer.
Up to now, several scientists have suggested distinct CNNs.

979-8-3503-0085-7/23/$31.00 ©2023 IEEE 329

Authorized licensed use limited to: MICROSOFT. Downloaded on December 28,2024 at 07:34:27 UTC from IEEE Xplore. Restrictions apply.
 Proceedings of the International Conference on Self Sustainable Artificial Intelligence Systems (ICSSAS 2023)
IEEE Xplore Part Number: CFP22DN7-ART; ISBN: 979-8-3503-0085-7

Titty et al. AI and It is mostly It minimizes the innovations in the world of medicine. Nevertheless, one
[32] among the major fascinating as well as difficult problems for
ML employed requirement for
in invasive doctors is making an appropriate prognosis of a disease's
biological processes. course. As a consequence, these approaches are able to find
research patterns as well as links among them in large datasets.
studies.

Nadauld MCED It can also The programs are IX. COMPARATIVE ANALYSIS
et al. [33]
test be not considered in
combined underserved The comparative analysis for the prediction of
with populations. cancer at an early stage can be described using two measures
traditional as follows.
clinical Accuracy analysis: The accuracy analysis of the
cancer considered state-of-the-art methods for the prediction of
diagnostic cancer at an early-stage cd using the table and graph below.
pathways. Table 2 Accuracy analysis
Chun et Supervis It doubles The tumor fraction
al. [34]
Methods Accuracy (%)
ed the enhances relative KNN 78.0
learning convention to before the SVM 79.08
approac al biopsy. DT 86.0
hes sensitivity NB 83.10
analysis in
DNN 91.19
breast
cancer
detection.
Klein et MCED The cancer It is not reflective
al. [35]
test signals are of efficiency in a
detected screening
across a population.
vast
diversity
of cancers.

VII. APPLICATIONS OF SMART PHONE IN CANCER

DIAGNOSIS AND THERAPY

A mobile app called SkinVision is accessible on

devices running Android 7.0 or later and iOS 13.0 or above.
Patients may use the app's partly free service to monitor and
examine skin patches for possible indications of skin cancer.
The app provides individuals with a thorough resource for
evaluating their signs and comprehending possible medical
concerns by acting as both a health library and a symptom
assessment platform.
Free mobile app offers tools as well as information Fig 2 Accuracy analysis
for recording signs, visits, and prescriptions connected to
cancer. It is made to be used by both patients as well
as medical professionals, rendering it a useful tool for Precision analysis: Using the table and graph
anybody impacted by cancer. On the basis of the most recent below, perform a precision analysis of the approaches that are
clinical data, the app gives doctors platforms for staging. thought to be state-of-the-art for predicting cancer at an early
stage.
VIII. PROBLEM DEFINITION
Early cancer evaluation and outlook are now Table 3 Precision analysis
essential to cancer studies since they may help with the Methods Precision (%)
following clinical care. A number of studies from
KNN 77.20
the bioinformatics and biomedical fields have studied the use
SVM 79.08
of ML and DL approaches due to the significance of
DT 85.79
categorizing cancer patients into low or high-risk categories.
Studies on cancer has seen a steady change NB 82.30
throughout the past few decades. Vast volumes of cancer DNN 90.57
information have been gathered and accessible to the medical
research sector thanks to the development of novel

979-8-3503-0085-7/23/$31.00 ©2023 IEEE 330

Authorized licensed use limited to: MICROSOFT. Downloaded on December 28,2024 at 07:34:27 UTC from IEEE Xplore. Restrictions apply.
 Proceedings of the International Conference on Self Sustainable Artificial Intelligence Systems (ICSSAS 2023)
IEEE Xplore Part Number: CFP22DN7-ART; ISBN: 979-8-3503-0085-7

explanation for their findings. Diagnostic methods need to be

understood, and doctors should be responsible.
Developing personalized methods: Methods for
mammographic breast cancer detection usually include the
assumption that the exhibiting features of individuals spread
uniformly. The method does not compensate for the fact that
various individuals may have particular traits that are unique
to them. By creating personalized methods that may take into
consideration the particular traits of every individual patient,
it can be feasible to increase the accuracy of mammographic
breast cancer detection.
Interpreting the hybrid technique: The hybrid
technique, like various techniques, combines machine
learning methods and the expertise of subject matter
expertise. Although there exists a chance that this tactic may
work, the method's decisions are not always clear-cut. One
strategy for enhancing confidence in the method's decisions
is to develop ways for understanding the hybrid technique.
Limitation of vast and diverse datasets: Deep
Fig 3 Precision analysis learning methods need datasets that are both huge and varied
in order to be taught efficiently. However, the availability of
these datasets is constrained, which might make it more
X. OPEN CHALLENGES WITH DISTINCT METHODS challenging to create precise deep learning methods for the
identification as well as evaluation of skin cancer. A possible
There exists a problem with appropriate datasets remedy to this issue might be the creation of datasets that are
that contain the necessary quantity of information for putting both broader in breadth and more diverse in their makeup.
into several ML as well as deep learning methods. Additional
image processing needs to be conducted more accurately for XI. SUMMARY OF THE COMPARATIVE ANALYSIS
improved classification outcomes for fresh information.
Clinical challenges: When applied to different In summary, current research on the application of
patient populations, AI methods may not perform as ML and DL methods for cancer detection has yielded
effectively. This could be challenging in cancer imaging, encouraging findings. According to the findings of this recent
because treatment outcomes and illness presentation differ research, these methods are capable of correctly diagnosing a
greatly. In terms of the accessibility of AI methods, number of cancer forms, involving lung cancer, breast
sophisticated AI methods could make it challenging for cancer, and oral cancer. To test their accuracy as well
doctors to understand how they arrived at a detection or as enhance their effectiveness, further study is required as
treatment recommendation. Additionally, this may make it these methods are yet in the early phases of evolution.
challenging to accept and use model outcome in clinical Notwithstanding this, employing DL and ML methods for
decision-making. AI methods should be effectively cancer detection has the possibility to enhance accuracy and
incorporated into healthcare workflows without interfering efficiency, making it an attractive field of research. It is
with present procedures. This could entail challenging tasks necessary to undertake further study on both the hybrid
like integrating imaging exams or changing test schedules. methods, which employ numerous methods in addition to just
When applying AI in clinical practice, concerns about data one to boost accuracy, and the effectiveness of these
security, privacy, and liability arise. Clinical use of AI methods in order to discover possible barriers to their
models and regulatory authorization might be challenging. deployment in clinical practice.
Addressing ethical considerations: Concerns about
patient privacy are moral. Imaging, genetics, and medical
history are needed for machine learning methods. Patients REFERENCES
might be harmed by security breaches or unauthorized access.
To fix this, software developers should restrict [1] Afshar P., Mohammadi A., Plataniotis K. N., Oikonomou A., and
information breaches and conceal patient information. In Benali H., "From handcrafted to deep-learning-based cancer
radiomics: Challenges and opportunities", IEEE Signal Process. Mag.,
order to opt out, patients must be informed about how their vol. 36, no. 4, pp. 132-160, Jul. 2019.
information will be utilized. Bias represents another ethical [2] Shakir H., Deng Y., Rasheed H., and Khan T. M. R., "Radiomics based
issue with cancer prediction machine learning algorithms. To likelihood functions for cancer diagnosis", Sci. Rep., vol. 9, no. 1, p.
lessen this risk, software programmers should train their 9501, Dec. 2019.
methods using data sets from patient populations. [3] Koo M. M., "Presenting symptoms of cancer and stage at diagnosis:
Professionals and patients can better understand diagnostics Evidence from a cross-sectional, population-based study", Lancet
Oncol., vol. 21, no. 1, pp. 73-79, Jan. 2020.
and information analysis thanks to simple methods. Defect
[4] Ferlay J., Soerjomataram I., Dikshit R., Eser S., Mathers C., Rebelo
detection may be challenging due to algorithmic M., Parkin D. M., Forman D., and Bray F., "Cancer incidence and
misdiagnosis, such as in cases of brain tumors. Designers of mortality worldwide: Sources, methods and major patterns in
algorithms should bear responsibility and provide an GLOBOCAN 2012", Int. J. Cancer, vol. 136, no. 5, pp. E359-E386,
Mar. 2015.

979-8-3503-0085-7/23/$31.00 ©2023 IEEE 331

Authorized licensed use limited to: MICROSOFT. Downloaded on December 28,2024 at 07:34:27 UTC from IEEE Xplore. Restrictions apply.
 Proceedings of the International Conference on Self Sustainable Artificial Intelligence Systems (ICSSAS 2023)
IEEE Xplore Part Number: CFP22DN7-ART; ISBN: 979-8-3503-0085-7

[5] Singh R. K. and Sivabalakrishnan M., "Feature selection of gene diagnostic resolution following multi-cancer early detection testing",
expression data for cancer classification: A review", Proc. Comput. Journal of Clinical Oncology, vol. 41, June 2023.
Sci., vol. 50, pp. 52-57, Jan. 2015. [25] Roshan Gangurde, Vishal Jagota, Mohammad Shahbaz Khan, Viji
[6] Bray F., Ferlay J., Soerjomataram I., Siegel R. L., Torre L. A., and Siva Sakthi, Udaya Mouni Boppana, Bernard Osei, and Kakarla Hari
Jemal A., "Global cancer statistics 2018: GLOBOCAN estimates of Kishore, "Developing an Efficient Cancer Detection and Prediction
incidence and mortality worldwide for 36 cancers in 185 countries", Tool Using Convolution Neural Network Integrated with Neural
CA, Cancer J. Clin., vol. 68, no. 6, pp. 394-424, 2018. Pattern Recognition", Biomed Res Int, January 2023.
[7] Bang O. Y., Chung J.-W., Lee M. J., Seo W.-K., Kim G.-M., and Ahn [26] Konstantina Kourou, Themis Exarchos P., Konstantinos Exarchos P.,
M.-J., "Cancer-related stroke: An emerging subtype of ischemic Michalis Karamouzis V., and Dimitrios Fotiadis I., "Machine learning
stroke with unique pathomechanisms", J. Stroke, vol. 22, no. 1, pp. 1- applications in cancer prognosis and prediction", Computational and
10, Jan. 2020. Structural Biotechnology Journal, vol. 13, pp. 8-17, 2015.
[8] Menéndez L. Á., de Cos Juez F. J., Lasheras F. S., and Riesgo J. A. Á., [27] Benjamin Hunter, Sumeet Hindocha, and Richard Lee, "The Role of
"Artificial neural networks applied to cancer detection in a breast Artificial Intelligence in Early Cancer Diagnosis", Cancers, vol. 14,
screening programme", Math. Comput. Model., vol. 52, nos. 7-8, pp. no. 6, March 2022.
983-991, Oct. 2010. [28] Farhana Shaikh and Rao D.S., "Prediction of Cancer Disease using
[9] Kim D. J., Rockhill B., and Colditz G. A., "Validation of the Harvard Machine learning Approach", Materials Today: Proceedings, vol. 50,
cancer risk index: A prediction tool for individual cancer risk", J. Clin. no. 3, April 2021.
Epidemiol., vol. 57, no. 4, pp. 332-340, 2012. [29] Abdu Rehaman Pasha Syed, Rahul Anbalagan, Anagha Setlur S.,
[10] Raoof S, Kennedy CJ, and Wallach DA, "Molecular cancer screening: Chandrashekar Karunakaran, Jyoti Shetty, Jitendra Kumar and Vidya
in search of evidence", Nat Med, vol. 27, no. 7, pp. 1139-1142, 2021. Niranjan, "Implementation of ensemble machine learning algorithms
[11] Cohen JD, Li L, and Wang Y, "Detection and localization of on exome datasets for predicting early diagnosis of cancers", BMC
surgically resectable cancers with a multi-analyte blood test", Science, Bioinformatics, vol. 23, November 2022.
vol. 359, no. 6378, pp. 926-930, 2018. [30] Somit Jain, Dharmik Naicker, Ritu Raj, Vedanshu Patel, Yuh-Chung
[12] Telonis AG, Magee R, and Loher P, "Knowledge about the presence Hu, Kathiravan Srinivasan and Chun-Ping Jen, "Computational
or absence of miRNA isoforms (isomiRs) can successfully Intelligence in Cancer Diagnostics: A Contemporary Review of Smart
discriminate amongst 32 TCGA cancer types", Nucleic Acids Res, Phone Apps, Current Problems, and Future Research Potentials",
vol. 45, no. 6, pp. 2973-2985, 2017. Diagnostics, vol. 13, April 2023.
[13] Lopez-Rincon A, Martinez-Archundia M, and Martinez-Ruiz GU, [31] Premalatha D, Niveditha N, Poornima Vasanth V, Priyanka G N, and
"Automatic discovery of 100-miRNA signature for cancer Niveditha K B, "Cancer prediction system using data mining
classification using ensemble feature selection", BMC techniques", International Journal of Computer Engineering and
Bioinformatics, vol., no. 1, 2019. Sciences Research, vol. 6, no. 6, 2019.
[14] Demir IE, Mota Reyes C, and Alrawashdeh W, "Future directions in [32] Austin Titty Philip, Shashank Pal, and Aarohi Verma, "A study of
preclinical and translational cancer neuroscience research", Nat cancer prediction using Neural Network", International Research
Cancer, vol. 1, pp. 1027-1031, 2021. Journal of Engineering and Technology (IRJET), vol. 7, no. 1,
[15] Hoadley KA, Yau C, and Hinoue T, "Cell-of-origin patterns dominate November 2020.
the molecular classification of 10,000 tumors from 33 types of [33] Lincoln Nadauld D., Charles McDonnell H., Tomasz Beer M.,
cancer", Cell, vol. 173, no. 2, pp. 291-304, 2018. Minetta Liu C., Eric Klein A., Andrew Hudnut, Richard Whittington
[16] Shakir H., Khan T., Rasheed H. and Deng Y., "Radiomics Based A., Bruce Taylor, Geoffrey Oxnard R., Jafi Lipson, Margarita
Bayesian Inversion Method for Prediction of Cancer and Pathological Lopatin, Rita Shaknovich, Karen Chung C., Eric Fung T., Deborah
Stage," in IEEE Journal of Translational Engineering in Health and Schrag and Catherine Marinac R., "The PATHFINDER Study:
Medicine, vol. 9, pp. 1-8, 2021. Assessment of the Implementation of an Investigational Multi-Cancer
[17] Muhammad Hammad Waseem, Malik Sajjad Ahmed Nadeem, Assad Early Detection Test into Clinical Practice", Cancers, vol. 13, July
Abbas, Aliya Shaheen, Wajid Aziz, Adeel Anjum, Umar Manzoor, 2021.
Muhammad a. Balubaid, and Seong-o shim "On the Feature Selection [34] Ka-Chun Wong, Junyi Chen, Jiao Zhang, Jiecong Lin, Shankai Yan,
Methods and Reject Option Classifiers for Robust Cancer Prediction," Shxiong Zhang, Xiangtao Li, Cheng Liang, Chengbin Peng, Qiuzhen
in IEEE Access, vol. 7, pp. 141072-141082, 2019. Lin, Sam Kwong, and Jun Yu, "Early Cancer Detection from
[18] Raweh A. A., Nassef M. and Badr A., "A Hybridized Feature Multianalyte Blood Test Results", iScience, vol. 15, pp. 332-341, May
Selection and Extraction Approach for Enhancing Cancer Prediction 2019.
Based on DNA Methylation," in IEEE Access, vol. 6, pp. 15212- [35] Klein E.A., Richards D., Cohn A., Tummala M., Lapham R.,
15223, 2018. Cosgrove D., Chung G., Clement J., Gao J., Hunkapiller N., Jamshidi
[19] Huitao Qi, Shuangbo Xie, Yanli Chen, Chengqian Wang, Tingting A., Kurtzman K.N., Seiden M.V., Swanton C., and Liu M.C.,
Wang, Bin Sun, and Mingxu Sun, "Prediction Methods of Common "Clinical validation of a targeted methylation-based multi-cancer
Cancers in China Using PCA-ANN and DBN-ELM-BP," in IEEE early detection test using an independent validation set", vol. 32, no.
Access, vol. 10, pp. 113397-113409, 2022. 9, pp. 1167-1177, September 2021.
[20] Jha S. K., Joo G. -S., Ra G. -S., Lee H. H. and Kim Y. -S.,
"Development of PCR Microchip for Early Cancer Risk Prediction,"
in IEEE Sensors Journal, vol. 11, no. 9, pp. 2065-2070, Sept. 2011.
[21] Swarna Ganesh, Krishnan Venkatakrishnan, and Bo Tan, "Early
detection and prediction of cancer metastasis – Unravelling metastasis
initiating cell as a dynamic marker using self- functionalized
nanosensors", Sensors and Actuators B: Chemical, vol. 361, June
2022.
[22] Charles McDonnell H., Andrew Hudnut G., Deepti Behl, and Roger
Ang, "Diagnostic workup following a multicancer early detection test
with a cancer signal origin prediction", Journal of Clinical Oncology,
vol. 40, June 2022.
[23] Juntaro Matsuzaki, Ken Kato, Kenta Oono, and Naoto Tsuchiya,
"Prediction of tissue-of-origin of early-stage cancers using serum
miRNomes", JNCI Cancer Spectrum, vol. 7, no. 1, November 2022.
[24] Eric Klein A., Charles McDonnell H., Lincoln Nadauld, and Christina
Dilaveri A., "Clinical evaluation of cancer signal origin prediction and

979-8-3503-0085-7/23/$31.00 ©2023 IEEE 332

Authorized licensed use limited to: MICROSOFT. Downloaded on December 28,2024 at 07:34:27 UTC from IEEE Xplore. Restrictions apply.