Application of Machine Learning in Healthcare and Medicine: A Review

— This extensive literature review investigates the integration of Machine Learning (ML) into the healthcare sector, uncovering its potential, challenges, and strategic resolutions. The main objective is to comprehensively explore how ML is incorporated into medical practices, demonstrate its impact, and provide relevant solutions. The research motivation stems from the necessity to comprehend the convergence of ML and healthcare services, given its intricate implications. Through meticulous analysis of existing research, this method elucidates the broad spectrum of ML applications in disease prediction and personalized treatment. The research's precision lies in dissecting methodologies, scrutinizing studies, and extrapolating critical insights. The article establishes that ML has succeeded in various aspects of medical care. In certain studies, ML algorithms, especially Convolutional Neural Networks (CNNs), have achieved high accuracy in diagnosing diseases such as lung cancer, colorectal cancer, brain tumors, and breast tumors. Apart from CNNs, other algorithms like SVM, RF, k-NN, and DT have also proven effective. Evaluations based on accuracy and F1-score indicate satisfactory results, with some studies exceeding 90% accuracy. This principal finding underscores the impressive accuracy of ML algorithms in diagnosing diverse medical conditions. This outcome signifies the transformative potential of ML in reshaping conventional diagnostic techniques. Discussions revolve around challenges like data quality, security risks, potential misinterpretations, and obstacles in integrating ML into clinical realms. To mitigate these, multifaceted solutions are proposed, encompassing standardized data formats, robust encryption


INTRODUCTION
Machine Learning (ML) and the field of medical care are fundamentally two separate realms.However, in recent times, developments in the field of Artificial Intelligence (AI), particularly in ML, have opened up intriguing new opportunities in medical treatment [1].The dynamic intersection between ML and the domain of medical care has captured the attention of researchers and healthcare practitioners alike, and has also triggered a paradigm shift in the approach towards medical treatment [2], [3].ML, as a component of AI, encompasses algorithms that enable computers to learn patterns from data, adapt, and make predictions or decisions without explicit programming [4]- [6].ML has rapidly evolved from a theoretical concept into a tangible force that holds the promise of revolutionizing how diseases are diagnosed, treated, and managed [7]- [9].
The integration of ML into medical practices is a response to the skyrocketing challenges faced by the healthcare industry.As medical data explodes in volume and complexity, traditional approaches to diagnosis and treatment are being put to the test.The effectiveness of medical decisions relies on assimilating and interpreting a wide range of patient data, including medical histories, medical imaging results, genomic information, and clinical records.ML has the potential to sift through these intricate data patterns, unveil hidden correlations, and extract insights that can guide more accurate diagnoses and tailored treatment plans [10], [11].
The crux of this topic lies not only in its technological implementation but also in its profound human potential.Successful integration of ML into medical care can yield faster diagnoses, reduce medical errors, and optimize resource allocation.This is crucial in addressing the increasing demand for high-quality healthcare services while contending with resource limitations and time constraints.This review article aims to elucidate the intricate relationship between ML and medical treatment by delving into the mechanisms through which ML algorithms operate in a medical context.It also endeavors to explore various fields within medical care where ML techniques have shown promising results through systematic examination of existing literature, case studies, and ongoing research projects.The article also highlights examples where ML has demonstrated transformative potential in disease diagnosis [12]- [15], prognosis [16]- [18], personalized treatment [19]- [21], drug discovery [22], [23], and patient management [24], [25], Furthermore, the article seeks to provide an in-depth analysis of the challenges that invariably accompany the integration of ML into medical treatment.These challenges encompass a range of issues, from data privacy concerns and ethical considerations to technical barriers and the need for interpretable models.From these issues, this article also attempts to discuss potential solutions or steps that can be taken to address the existing problems [26]- [28].
Therefore, the discussions in this article are expected to encourage the advancement of medical practices through the integration of AI and wise human values.

A. Introduction to ML in Medical Treatment
ML is a branch of artificial intelligence that teaches computers to learn from data and make decisions based on patterns found in that data.In the context of medical treatment, ML enables computers to process medical data, identify health patterns, and make predictions without the need for explicit programming.
The fundamental concept behind ML involves several components such as models, training, and evaluation [29].A model is a mathematical representation of the relationships between variables in the data.This model can take the form of mathematical functions or structures that depict how variables influence each other.The primary goal of creating a model is to enable the computer to recognize hidden patterns or rules within the data.A good model will be capable of accurately representing the relationships between variables.
The training process is at the core of ML.During training, the model is provided with knowledge and is trained using data to identify patterns and make accurate predictions.Once the model is trained, the next step is to measure how well the model performs on unseen data (test or validation data).Evaluation aims to understand to what extent the model can generalize the patterns learned during training.Common evaluation metrics include accuracy [30], [31], precision and recall [32]- [34], F1-score [35]- [37], confusion matrix [38], ROC, Mean Absolute Error (MAE) [39], and others.These metrics help gauge the model's ability to make correct predictions and avoid errors that could have serious implications in medical treatment.These three fundamental concepts work together to create an effective model that understands medical data and makes accurate predictions.It's important to remember that the quality of training data significantly impacts the model's performance.Models trained with high-quality data tend to have better capabilities in recognizing patterns and providing more accurate prediction outcomes.The application of ML in the medical field is illustrated in Fig. 1.

B. Categories of ML Algorithms
ML algorithms can be categorized into several groups based on the type of learning they utilize, namely Supervised Learning (SL), Unsupervised Learning (UL), and Reinforcement Learning (RL) [40].SL is a type of ML that trains models using data containing patient examples and corresponding labels, such as diagnoses or treatment outcomes.Popular algorithms in this category include Random Forest (RF) [41], Support Vector Machine (SVM), and Neural Networks.
UL is a type of ML that trains models by identifying patterns in unlabeled data.It is suitable for grouping patients into specific categories based on shared characteristics.Popular algorithms in this category include K-Means Clustering and Hierarchical Clustering.On the other hand, RL is a ML approach in which algorithms learn through repeated interactions with their environment.RL algorithms, often referred to as "agents," learn to take actions that optimize a goal, such as maximizing rewards provided by the environment.Agents receive feedback in the form of rewards or punishments after each action taken, and their objective is to learn the best decisions based on this feedback.RL is often used in applications like robot control, computer games, and resource optimization.While RL is less commonly used directly in medical treatment, its concept can be applied in the development of optimal treatment planning algorithms.A comparison illustration of SL, UL, and RL is shown in Fig. 2.
In addition to the core categories mentioned above, there are several other specialized categories of ML algorithms.These include semi-supervised learning [63], [64], where algorithms leverage both labeled and unlabeled data; transfer learning, which involves reusing a pre-trained model for a related task; deep learning [65]- [68], utilizing neural networks with multiple layers for complex pattern recognition; ensemble learning [69]- [71], combining multiple models to improve predictive accuracy; anomaly detection [72], [73], identifying rare or abnormal instances; NLP algorithms [74], enabling machines to understand human language; and time series forecasting algorithms [75], [76], predicting future values based on historical data patterns.

A. Disease Progression Prediction
Predicting disease progression is one field where ML plays a crucial role.ML algorithms can analyze and identify patterns related to disease progression by leveraging patient data such as medical history, symptoms, laboratory tests, and medical images.A concrete example is the use of ML to predict the risk of diabetes in patients by analyzing data like blood glucose levels, body mass index, and family history.ML can identify significant risk factors and provide more accurate predictions of diabetes risk compared to traditional methods.Thus, the use of ML in predicting disease progression offers the advantage of guiding early interventions and preventing more severe complications.

B. Personalization of Treatment and Therapy
Personalized treatment and therapy are crucial aspects of modern medical care [77].Each patient possesses unique characteristics that influence their response to treatment.In this regard, ML can play a significant role in assisting doctors to design tailored treatment plans according to individual needs.For instance, in cancer treatment, ML can analyze patients' genetic data and responses to previous therapies to predict the most likely successful treatment.This avoids a one-size-fits-all approach and ensures that each patient receives the most appropriate care for their condition.

IV. LITERATURE STUDY OF ML APPLICATION IN MEDICINE
Effective healthcare data management is crucial for providing quality healthcare services and conducting meaningful research.ML plays a pivotal role in processing and comprehending large volumes of health data, often referred to as "big data".ML algorithms can identify patterns, trends, and relationships within extensive datasets that might be overlooked by human analysis [78]- [80].This empowers healthcare providers and researchers to extract valuable insights, such as identifying risk factors, tracking disease progression, and evaluating treatment outcomes.ML techniques like clustering and classification enable the organization and categorization of patient data, facilitating more accurate diagnoses and tailored treatment plans [81]- [83].
In the realm of medical diagnostics, ML has facilitated the development of automated systems capable of diagnosing diseases through the analysis of medical images like MRI and CT scans [84], [85].ML's impact extends to personalized care, enabling a more individualized approach by leveraging patient data and clinical histories to develop predictive models that respond specifically to each patient's needs.Consequently, ML leads to more efficient and effective treatments.The field of genomics is also influenced by ML, with its ability to analyze complex genomic data to identify genetic patterns associated with diseases or responses to medications, driving the development of targeted and precise treatments.
On the research front, ML aids in analyzing data from large-scale clinical studies more swiftly and accurately, enabling the identification of trends, risk factors, and therapy responses.Particularly, ML-based patient monitoring algorithms can detect subtle changes in patient data in realtime, assisting medical teams in responding to conditions that require immediate action.Through NLP, ML also enables the analysis of unstructured clinical data, such as medical records and radiology reports, to support better clinical decisionmaking [86]- [88].
In the pursuit of new drug discovery, ML assists in predicting drug potential based on molecular structure and biological interactions, expediting the drug discovery and Furizal, Application of Machine Learning in Healthcare and Medicine: A Review development process [89], [90].However, certain literature also raises ethical and security concerns related to the use of ML in medical contexts, including patient data privacy considerations, ML model interpretation, and the ethical implications of integrating medical decision-making with algorithms.Several literature studies on the application of ML in the medical field are presented in Table I.Table I represents a collection of research studies evaluating the use of ML techniques in various medical contexts, ranging from disease diagnosis to cancer detection.Each row in the table represents a specific research study and includes information about the identified disease, types of data used, data sources, applied ML algorithms, evaluation methods, achieved results, and the year of the study.
Through the compilation of research studies presented in the table, a profound conclusion can be drawn regarding the role and impact of ML in the medical field.These studies have provided crucial insights into how ML can be employed for disease diagnosis, medical condition classification, and enhancement of clinical decision-making.Upon comparing these studies, certain findings and patterns stand out, while challenges and opportunities become evident.
From the perspective of disease diagnosis, studies [42] and [43] focusing on leukemia (ALL) demonstrate that ML can address the complexity of medical data analysis with a relatively high accuracy, namely 95.6% and 93.84%.The use of SVM and other algorithms in analyzing data patterns enables the identification of disease symptoms with consistent outcomes.A similar trend can be observed in studies [44], [45], [48], where the application of SVM, k-NN, RF, LR, and CNN algorithms showcases the capability of ML in classifying various diseases, spanning from white blood cells to cardiac arrhythmias and brain and breast tumors, achieving accuracy rates ranging from 80.8% to 92.8%.
When involving medical videos, study [46] has demonstrated that ML can yield high-accuracy results in identifying colorectal cancer with an accuracy of 90.28%.This outcome highlights ML's significant potential as a valuable tool in accurately interpreting and classifying medical videos.Similarly, when dealing with medical images [48], [50], [51], [55], [56], [61], ML has also proven to deliver commendable outcomes with accuracies ranging from 83.64% to 99.86%.

A. Data Quality and Quantity
The primary challenge in adopting ML techniques in the medical field is the complexity and variability of medical data generated from various sources and healthcare information systems.Medical data is often distributed across diverse formats, including clinical records, medical images, genomic data, and more [91], [92].This challenge encompasses difficulties in integrating and processing data with different structures, formats, and languages [93].Additionally, medical data is susceptible to noise, recording errors, and variations in interpretation by healthcare practitioners, which can impact the quality and accuracy of the resulting ML models.Limited and fragmented data availability can also affect the model's ability to generate generalized and valid predictions across various medical scenarios.
To address these challenges, a holistic approach involving improved medical data integration and data quality enhancement is necessary.Firstly, standardizing the format and structure of medical data can help address data diversity.The use of standards such as Health Level Seven International (HL7) for data exchange and formats like Digital Imaging and Communications in Medicine (DICOM) for image-based medical data can reduce integration barriers [94]- [98].Additionally, technologies like NLP can be employed to handle unstructured data, such as medical records or radiology reports, transforming them into information that can be processed by ML algorithms [99]- [101].This approach can be bolstered by the implementation of integrated, cloud-based data management systems, enabling efficient access and exchange of medical data across healthcare institutions.With these solutions in place, the main challenges in harnessing ML for medical purposes can be overcome, unlocking the significant potential of ML in healthcare treatment and diagnostics.

B. Data Privacy and Security
The challenge of ensuring data privacy and security is a critical concern when implementing ML in the medical domain.Medical data contains sensitive and confidential information about patients, including their health conditions, treatment histories, and personal identifiers [102]- [105].As ML techniques involve processing and analyzing this data, there is a risk of unauthorized access, data breaches, and potential misuse of patient information.Moreover, the increasing adoption of cloud-based solutions for data storage and processing introduces additional complexities in safeguarding data against potential cyber threats and vulnerabilities.
To address the challenge of data privacy and security, stringent measures must be put in place.Firstly, robust encryption techniques should be employed to secure data both at rest and during transmission.This helps protect patient information from being accessed by unauthorized parties.Secondly, the implementation of access controls and authentication mechanisms ensures that only authorized personnel can access sensitive medical data.Regular monitoring and auditing of data access can help identify any unusual activities promptly.Additionally, anonymization and de-identification techniques can be applied to remove personally identifiable information from datasets used for ML training, reducing the risk of re-identification.
Collaboration with cybersecurity experts and adherence to established industry standards, such as the Health Insurance Portability and Accountability Act (HIPAA) in the United States [106]- [108] or the General Data Protection Regulation (GDPR) in Europe [109], [110], can provide guidelines and best practices for ensuring data privacy and security in the context of ML in healthcare.By adopting these measures, healthcare organizations can maintain patient trust and ensure that data remains protected while benefiting from the advancements brought by ML technologies.

ISSN: 2715-5072 625
Furizal, Application of Machine Learning in Healthcare and Medicine: A Review

C. Misinterpretation
Misinterpretation of ML results is a significant challenge in the medical field, which can have profound implications for patient care and decision-making.ML models often operate as complex "black-boxes," making it difficult to understand the underlying factors that contribute to their predictions.This lack of interpretability can lead to difficulties in validating the reliability and accuracy of the model's outputs, especially in critical medical scenarios.Misinterpretation can occur when healthcare professionals either overly rely on ML predictions without understanding their limitations or misjudge the confidence level of a prediction, potentially leading to incorrect diagnoses or treatment plans.
To mitigate the challenge of misinterpretation, several strategies can be employed.Firstly, developing interpretable ML models is essential.Techniques such as feature importance analysis, SHAP (SHapley Additive exPlanations), and LIME (Local Interpretable Modelagnostic Explanations) can shed light on how the model arrived at a particular prediction by highlighting the most influential features [111], [112].Secondly, providing clinicians and medical practitioners with proper training in understanding and interpreting ML results is crucial.Healthcare professionals should be aware of the strengths and limitations of the models they are using and should be encouraged to critically assess the predictions in the context of their clinical expertise.Collaborative efforts between data scientists, clinicians, and domain experts can bridge the gap between technical understanding and medical practice, ensuring that ML results are used effectively and responsibly.Furthermore, transparency in model development and reporting, including documentation of the dataset used, preprocessing steps, and model architecture, can enhance accountability and facilitate peer review, aiding in the accurate interpretation of results.By addressing misinterpretation through a combination of model interpretability, education, and collaboration, the medical community can harness the power of ML while maintaining the highest standards of patient care and safety.

D. Clinical Acceptance
The main challenge in achieving clinical acceptance of ML technology in the medical field is to build confidence and trust among healthcare professionals in the effectiveness and reliability of ML models [113].Medical practitioners typically rely on established practices and scientific evidence, and integrating new technologies like ML can trigger uncertainty and resistance.Overcoming concerns related to accuracy [114], clinical validity, and the risk of errors arising from the interpretation or recommendations of ML models is crucial.
One key solution is close collaboration between data scientists, medical practitioners, and domain experts.Ensuring that ML models are based on relevant and representative data and applied in the appropriate medical context is a vital step in building clinical acceptance.Model development also needs to consider the understanding of medical practitioners about the algorithms and evaluation metrics used.Additionally, it's important to prioritize a transparent and interpretable approach in ML decisionmaking, so that medical practitioners can comprehend and feel confident in the outcomes and recommendations provided by the model.Proper education and training are also necessary to help healthcare professionals understand the added value offered by ML technology and how to integrate it safely and effectively into their daily clinical practice.Therefore, a collaborative and comprehensive approach involving medical and technological stakeholders will contribute to broader clinical acceptance of ML technology in the medical field.

E. Interoperability
Interoperability stands as a critical challenge in adopting ML technology in the medical field.Health data is often scattered across various systems, platforms, and different formats, making integration and exchange of data among healthcare entities challenging.The inability of systems and applications to communicate seamlessly can hinder ML's ability to harness comprehensive information from diverse data sources.This situation often leads to inefficiencies in data management and reduces the effectiveness of more holistic and accurate analyses.
To address interoperability challenges, a crucial step is to develop standardized data and exchange protocols that are uniform across the healthcare industry.Adopting standards like Fast Healthcare Interoperability Resources (FHIR) can enable consistent data exchange that can be interpreted by various systems [115]- [117].Furthermore, leveraging Application Programming Interfaces (APIs) can facilitate communication and data integration across different platforms [118], [119].Thus, collaboration and information exchange among healthcare institutions can be enhanced, supporting the effective and comprehensive application of ML in health data analysis.

F. Resource Constraints
Resource constraints pose a significant challenge in the adoption of ML in the medical domain.ML algorithms require substantial computational power and memory, especially for processing and analyzing large-scale medical datasets.Many healthcare facilities face limitations in terms of available hardware, software, and technical expertise, hindering the seamless implementation of ML solutions.These constraints can hinder the timely and efficient deployment of ML models, delaying the potential benefits they could bring to medical decision-making and patient care.
To address resource constraints, a combination of strategies can be employed.Cloud computing offers a solution by providing scalable and flexible resources ondemand, reducing the burden on local hardware infrastructure.Healthcare institutions can leverage cloud platforms to access powerful computational resources without investing heavily in physical hardware.Collaborating with technology partners or vendors specializing in healthcare-oriented ML solutions can also mitigate resource challenges [120].Such partnerships can provide healthcare professionals with access to cutting-edge algorithms and expertise, allowing them to focus on the medical aspects rather than the technical complexities.By strategically utilizing cloud resources and engaging with external expertise, healthcare facilities can overcome resource limitations and effectively harness the potential of ML for medical advancements.

G. Medical Ethics
The integration of ML in the medical field introduces complex ethical challenges.One of the main concerns is the potential impact on patient privacy and confidentiality [121]- [123].ML algorithms often require access to sensitive patient data, raising questions about data security, informed consent, and the risk of unauthorized access or breaches.Another challenge involves the transparency of ML algorithms' decision-making processes.
Addressing these challenges requires a multi-faceted approach.To ensure patient privacy, robust data protection measures should be implemented, such as data anonymization and encryption.Healthcare institutions should also prioritize obtaining informed consent from patients before their data is used for ML purposes, fostering transparency and respect for patient autonomy.Additionally, the development of interpretable and explainable ML models can enhance their ethical standing.By implementing these solutions, the ethical challenges associated with medical applications of ML can be effectively addressed, promoting responsible and patient-centered deployment of technology in healthcare.
All the challenges and solutions in implementing ML in healthcare and medicine in this article have been summarized in Fig. 3.

VI. CONCLUSSION
In conclusion, the comprehensive analysis of literature in the application of ML within the realm of healthcare and medicine reveals its remarkable potential and significant challenges.The studies discussed, particularly those involving disease diagnosis and medical image interpretation, underscore the substantial accuracy achieved by ML algorithms, with some even surpassing 90%.Notably, CNNs and other techniques like SVM, RF, k-NN, and DT play pivotal roles in achieving these impressive results.This demonstrates ML's transformative impact on medical practices, from enhancing disease detection to enabling precise medical image analysis.However, challenges persist, notably in ensuring data quality, managing complex datasets, and addressing variations that affect ML algorithm effectiveness.These obstacles underline the necessity of ongoing research and collaboration among multidisciplinary stakeholders, including medical professionals, data scientists, and technologists.Overcoming challenges requires standardized data formats, robust encryption for privacy, interpretability to build trust, comprehensive clinician training, and enhanced collaboration among stakeholders.

Fig. 1 .
Fig. 1.Illustration of applying ML in the medical field

Fig. 2 .
Fig. 2. Illustration of the comparison of working concepts of SL, UL, and RL in the medical field

Fig. 3 .
Fig. 3. Challenges and solutions in implementing ML in healthcare and medicine