The earliest publications on how AI enabled healthcare sector data were from 2010 (see Fig. 3). The greatest number of articles, 44, was published in 2020, and the lowest number of articles, 4, was published in 2012. Most of the articles were published between 2018 and 2021, thus signifying the recent interest in this research area.

Fig. 3.

Publications by year

(0.16MB).

Fig. 4 illustrates the distribution of the selected articles by database source. 64 articles were identified in the IEEE database, followed by 39 articles from the ACM Digital database, 34 articles from the Scopus database, 19 articles from the Emerald database, 16 articles from the Science Direct database, and only 8 articles from the AIS e-library (see Fig. 4).

Fig. 4.

Distribution of articles by database source

(0.15MB).

The research topic is organized into four dimensions: benefits, challenges, methodologies, and functionalities. Fig. 4 displays the quantity of articles published each year in each dimension. Thus, the aggregate number of articles published under benefits is (n=86); challenges (n=44); methodologies (n=98); and functionalities (n=92). The majority of the articles investigating how AI enabled the healthcare sector are related to benefits, methodologies, and functionalities (Fig. 5).

Fig. 5.

Research topics according to classification framework

(0.26MB).

Nevertheless, the number of studies focusing on AI challenges has been the lowest in the past ten years. The benefits of using AI within organizations is another topic of interest to many researchers and is the third most studied topic as per this classification framework. The method that organizations used to adopt and use AI was the second most studied topic as per the classification framework. Meanwhile, functionalities that AI provides to improve the performance of organizations have gained the most significant research attention in the past decade.

AI has several benefits to people, including automated decision-making (Sqalli and Al-Thani, 2019; Minz and Mahobiya, 2017; Ribbens et al., 2019; Seeböck et al., 2019; Strachna and Asan, 2020), patient monitoring, including elderly patient monitoring (Wu et al., 2019; Rahman et al., 2020; Comito et al., 2020; Khan et al., 2018; Wong et al.,2020; Mary et al.,2020), early diagnosis (Sasubilli et al.,2020; Jahan and Tripathi, 2021; Ladgham, 2013; Ribbens et al., 2019; Sakkos et al., 2021; Razaak et al.,2014) and process simplification (McGregor et al.,2020; Sqalli and Al-Thani, 2019; Deng et al.,2019).

The benefits to individuals are primarily derived from their collected medical data that are often heterogeneous, complex and not normalized. In addition, they are often large in volume and come in different formats. While this represents a great challenge for medical care practitioners, AI can be very effective in analyzing these types of big data and coming up with innovative solutions that are highly relevant and meaningful for health practitioners, ultimately benefiting the patients in their care, diagnosis and treatment options. Generally speaking, diagnosis- and treatment-related decisions consume both time and effort. AI provides a convenient solution to overcome this challenge, as it has great potential in generating automatic inferences with limited or no human intervention. Some studies even argue that AI can outperform humans in some medical situations related to radiology, cardiology and tumor detection (Kaur et al., 2021; Miller and Brown, 2018).

Sqalli and Al-Thani (2019) address the case of chronic diseases, stating that chronic diseases in particular are exhausting to the healthcare sector in terms of effort and cost because patients frequently visit their healthcare providers because they need continuous care. Some of these visits are unnecessary and cause a waste of time and resources. The authors find that health coaching, which is a healthcare strategy that consists of educating patients to improve their health-related behaviors, also helps in reducing the expenses of chronic patients. They propose a model that integrates health coaching with AI to help patients manage their chronic diseases more effectively by minimizing the number of unnecessary visits. Similar to the system suggested by Murray et al. (2019), the system consists of sensors that capture biometric data, AI models that generate insights about medical conditions and visual analytics tools that display relevant data in graphical and textual formats.

Organizations use AI applications and IT tools to reduce costs (Kaur et al., 2021; Sqalli and Al-Thani, 2019; Zhou et al., 2020; Roa et al., 2011), detect fraud (Dhieb et al.,2020; Rawte and Anuradha, 2015; Capelleveen et al.,2013), improve performance (Ladgham et al.,2013; Minz and Mahobiya, 2017; Maduri et al., 2020), and provide workflow assistance (Bernardini et al., 2021; Wu et al., 2019; Deng et al., 2019; Daltayanni et al.,2012).

For instance, Murray et al. (2019) highlight the difficulty of automatically extracting knowledge using clinical information systems, mainly due to the limited level of data normalization and integration. The authors introduce an artificial AI-empowered network (AI-KEN) to overcome these challenges. It consists of a network based on an AI system that automatically generates clinical notes, using predictive text options to reach normalization, over clinicians’ original notes to ensure data normalization and integration from multiple sources. AI-KEN operates using three main steps. First, it normalizes data using declared transformation rules. Then, it transforms data with validated ontologies to integrate them. Finally, it uses a data analytics model to extract valuable knowledge from the integrated data. Manifold research has introduced AI as a solution for similar complexities to reduce the consumption of resources in the healthcare sector (Sqalli and Al-Thani, 2019; Kaur et al., 2021; Zhou et al., 2020; Roa et al., 2011). Bohr and Memarzadeh state that implementing AI in healthcare can reduce the annual US healthcare costs by USD 150 billion in 2026 (Bohr and Memarzadeh, 2020). This reduction may be applied to other countries but there are limited quantitative evaluation data given the recency of the impact of AI in the healthcare sector.

AI can also assist the healthcare sector by detecting fraudulent claims. Health insurance fraud is a serious challenge in the insurance sector. According to the Federal Bureau of Investigation (FBI), insurance fraud causes an annual $80 billion loss in the USA (Austin, 2018). Moreover, 21% of bodily injury and 18% of personal injury claims that have been sealed with a full refund are fraudulent (Corum, 2017). Fraud detection becomes more challenging to detect with the continuous increase in data volume. Several researchers have attempted to integrate AI into the health insurance sector to detect fraudulent claims (Dhieb et al., 2020; Rawte and Anuradha, 2015; Capelleveen et al., 2013; Gangopadhyay and Chen, 2016; Anbarasi and Dhivya, 2017; Thorntonl, 2015; Ragde et al., 2016; Yang and Hwang, 2014; Bagde and Chaudhari, 2016). Rawte and Anuradha (2015) recommend a hybrid solution that combines clustering and classification to detect duplicated insurance claims. The authors mention that classification methods and clustering methods alone are not sufficient to detect duplicated claims. To overcome this challenge, the authors propose a two-step model. The first step is the claim clustering step performed using the evolving clustering method (ECM). The second step is based on the SVM algorithm classification step that uses the output of the first step as input. Ibrahim (2022) proposed AI based electronic health data recording system in short- and long-term hospitalization.

Dhieb et al. (2020) suggest an insurance system framework that uses extreme gradient boosting (XGBoost) to reduce human interaction, secure insurance activities, alert and inform risky customers, detect fraudulent claims, and reduce monetary loss for the insurance sector. After presenting the blockchain-based framework to enable secure transactions and data sharing among different interacting agents within the insurance network, they propose to employ the machine learning algorithm for the aforementioned insurance services and compare its performance with that of other state-of-the-art algorithms. The results reveal that when applied to an auto insurance dataset, XGboost achieved high performance gains compared to other existing learning algorithms. These are some examples of how recent AI advancements bring benefits to the organizations adopting the AI systems.

The entire healthcare sector that comprises the hospitals, insurance companies and government agencies also benefit from AI technologies as a whole. Since IT helps to collect, process, share and store patient data (Jumelle et al., 2014; Antoniou et al., 2018; Jaiman and Urovi, 2020; Bennett et al., 2011), building on this data AI can provide support towards professional training (Deist et al., 2020; Paranjape et al., 2019; Torner et al., 2019; Gupta et al., 2019), and such applications can save time (Sasubilli et al., 2020; Strachna and Asan, 2020; McGregor et al., 2020) and reduce resource consumption (Dhieb et al., 2020; Kaur et al., 2021; Kaur et al., 2021).

Extending the application of AI solutions to the field of healthcare brings countless advantages to this crucial field. It helps minimize resource consumption and reduce treatment expenses and time (Bohr and Memarzadeh, 2020). In addition, it accelerates the diagnosis time and thus the decision-making process, which has a major impact on treatment strategies and outcomes and might be lifesaving in critical situations. Data sharing in the health sector is important for individual well-being when followed by different doctors in different health institutions. It is also important for scientific research and progress. However, accessing medical data is often difficult due to regulatory and privacy challenges, according to Deist et al. (2020), who recommend the personal health train (PHT) system, which provides a privacy-by-design infrastructure connecting data sources while promoting the application of machine learning techniques to distributed data. Paranjape et al. (2019) encourage the use of AI for educational purposes and introduce a framework that includes AI into the medical education curriculum. Holistically speaking, AI can be beneficial across the entire healthcare sector.

There are obvious privacy issues related to accessing, editing, sharing and using patient data. Cloud computing and AI are often commonly used in different types of applications in the health sector. These systems collect, process, store, monitor and share health data (Sasubilli et al., 2020; Zerka et al., 2020; Peters et al., 2020; Jumelle et al., 2014). Despite the advantages of these systems, there are also some challenges such as security issues, privacy implications, cybersecurity and ethical issues. Hospitals and government agencies usually provide ethical procedures for collecting and sharing data. Permission is required from a government-approved authority to collect and use data, even for research purposes (Wang et al., 2020; Zhou et al., 2021; Bouguettaya et al., 2016). Other ethical issues associated with AI in the healthcare and broader sectors are inequality, unemployment, humanity, commitment to cause, regulatory approaches, behavioral biases, population biases and linking biases (Jameel et al., 2020). In order to reduce ethical issues in AI applications in the health area, there are studies on minimizing negative side effects, reward hacking, safe exploration and robustness (Jameel et al., 2020; Bryson et al., 2018; Zerka et al., 2020; Esmaeilzadeh et al., 2020). Some machine learning algorithms are used for the early prediction, treatment and diagnosis of diseases, and they can make decisions or support doctors’ decisions. Governments have expressed their concerns about these automated processes with respect to protecting patients’ rights. Such concerns have led to several regulations in data collection, processing, technology usage and quality of such data and collection and analysis methodologies. In addition, researchers who are working in the healthcare domain should have a careful focus on data quality, testing of data and documentation (Jumelle et al., 2014; Shaban-Nejad et al., 2021; Esmaeilzadeh et al., 2020) before AI applications can use them.

Some AI methods require a large volume of data to process. It is sometimes difficult to collect data, especially patient data, because of ethical implications of such data. The application of some classification and clustering algorithms on a very limited amount of data may yield very good accuracy; however, it might not be realistic and applicable (Patii and Iyer, 2017; Murray et al., 2019; Bennett et al., 2011). The collected data require preprocessing to be used in AI techniques. In particular, text data require considerable natural language processing before use. Different types of data, such as text, numeric, image and video, sometimes must be integrated using the same algorithm, which is one of the most difficult challenges in medical data processing (Patii and Iyer, 2017; Murray et al., 2019; Bennett et al., 2011). Medical data can be collected from various sources and formats, such as medical images, 3D video sequences, photographs, and numeric data. Collecting clean, robust and efficient data is a challenge in healthcare data analysis.

Machine learning, natural language processing and expert systems use medical data as an input to process and create models to support medical decisions based on health systems. Most of the applications of AI in health systems are related to diagnosis. False decisions in automated diagnosis may have very harmful results. Collected data from hospitals are sometimes not of sufficient quality or simply inaccurate. Data errors are among the top challenges in medical data processing via AI (Nikolaev et al., 2021; Ling et al., 2013; Hasan et al., 2017; Goldberg et al., 2010). Another challenge is decision errors made by machine learning algorithms. Sometimes the applied algorithm is not suitable for the given data, or the data are not sufficiently reliable to be used in classification algorithms such as neural networks, decision trees and Bayesian networks. Several studies have demonstrated possible decision-making problems in the health domain and their solutions (Madanan et al., 2021; Aljaaf et al., 2015; Srivastava and Rossi, 2019; Poletti et al., 2013; Kusano et al., 2017). Currently, a considerable number of AI and IoT devices and software are used in the healthcare sector. However, not all of them are automated: doctors make the final decision, and this interplay between the healthcare practitioners and AI models may end up with false diagnoses and treatment results (Sqalli and Al-Thani, 2019; Lim and Thuemmler, 2015; Mohr et al., 2011; Tsang et al., 2020).

There are several data types in AI applications in the healthcare sector, such as text, numeric data, images, videos and audio. Most of the time, data collected from patients are raw data and need processing before applying machine learning or similar algorithms. For example, quantitative electroencephalography (QEEG) is a signal type that is preprocessed, and several features are extracted for use in the high-level learning process (Ciprian et al., 2020). The quality of ultrasound video sequences is another important parameter of automated early diagnosis. Video encoding and linear regression methods increase the quality of the video (Panayides et al., 2018).

Machine learning algorithms can be used to understand personal health situations. Random forest and other rule-based algorithms are used in analyzing health patterns (Sasubilli et al., 2020). Truth maintenance system (TMS) algorithms predict features from large datasets and apply classification. TMS algorithms give more accurate results than other algorithms in human health pattern recognition. Several algorithms have also been developed for the early diagnosis and treatment of kidney disease (Bernardini et al., 2021). Machine learning algorithms provide monitoring and analysis of the estimated glomerular filtration rate (eGFR), which is an important parameter in the treatment of kidney disease. Another important data application in medicine is magnetic resonance imaging (MRI), which can be applied to image processing, computer vision and machine learning techniques for diagnosis (Jahan & Tripathi, 2021; Ladgham et al., 2013; Minz and Mahobiya, 2017; Nikolaev et al., 2021; Ribbens et al., 2014; Zhong et al., 2020; Ling et al., 2013; Amrane et al., 2018). Lung and breast cancers are two of the most frequent cancers that cause death. Eight percent of women are faced with breast cancer during their life. Machine learning algorithms have a very high success ratio for automated breast cancer detection (Amrane et al., 2018). A naïve Bayes classifier (NBC) and k-nearest neighbor (KNN) algorithm are used to classify the breast as either normal or abnormal. KNN gives a very high accuracy of 97.51%.

Likewise, Ribbens et al. (2014) proposed a new unsupervised segmentation and classification approach for brain MRI images. First, MRI was applied to segmentation to retrieve the feature vector. If clinical experts are experienced in record taking, supervised training can be applied; otherwise, unsupervised training can be applied to classify the brain as normal or requiring further investigation. Probabilistic classification methods yield very promising results in brain MRI images. However, it is difficult to obtain real-time feedback for asthma and eHealth management. Several well-known algorithms, especially NBC and logistic regression supervised classification algorithms, have been developed for early warning systems (Tsang et al., 2020). Yu (2022) and Causey (2022) proposed several object detections, segmentation and tracking methods to recognize tumors in medical images. Results from the use of AI are very promising especially with CT images.

Machine learning algorithms and the Internet of Things (IoT) were also used in several applications of the prediction and data collection of COVID-19 (Dharani and Krishnan, 2021; Hossen and Armoker, 2020; Gazzah and Bencharef, 2020; Elbasi et al., 2021). Probabilistic-based machine learning algorithms such as the Bayesian network and NBC were used to predict the possibility of a patient having COVID-19 (Dharani and Krishnan, 2021). Artificial neural network algorithms produce very promising results in the classification of patient data as virus or no virus. The IoT collects data from crowds using several sensors, such as temperature, distance and the use of masks. Collected data are used as an input to classification algorithms, which provide safe or unsafe crowds in indoor and outdoor public areas.

Expert system techniques also have several applications in health classification, decision-making and patient support systems (Sqalli and Al-Thani, 2019; Yu and Zhou, 2021; Chen et al., 2018; Bernardini et al., 2021; Gandhi et al., 2019; Reddy and Chavan, 2021; Albayrak et al., 2018). The management of chronic diseases is becoming important because of increases in patient visits. These patients need more frequent doctor and laboratory visits, which are time-consuming and costly. Sqalli and Al-Thani (2019) proposed expert system-based models to manage their decision support system. This automated health coaching system collects patient data, performs an analysis, retrieves doctor consultations and gives feedback to patients.

Medical image processing uses several types of images of the human body for early diagnosis and treatment. Computed tomography (CT) and magnetic resonance imaging (MRI) are frequently used in medical image processing research, and several applications already exist in the sector. Doctors, engineers and radiologists work together to analyze medical imaging (Nikolaev et al., 2021; Zhong et al., 2020; Alamri et al., 2010). Medical image processing takes images as an input and applies image processing techniques to understand images, such as resizing, cropping, rotation and editing. Nikolaev et al. (2021) developed a new algorithm using MRI images to diagnose breast cancer. Data preprocessing, segmentation, feature extraction and classification represent the sequence of methods used in breast cancer detection using AI techniques (Zhou et al., 2018; Harinishree et al., 2021; Amrane et al., 2021; Gayathri and Sumathi, 2021; Charan et al., 2018; Osareh and Shadgar, 2018). In addition to image processing, video processing is used in the health sector. Video cameras capture frames from inside the human body to collect more information, which means that in the health sector and AI, early and right decisions can be made. Video capturing also helps doctors in surgery. Sometimes a 2D image is not sufficient to make the right decision. In that case, 3D images can be used, such as for the brain or other organs (Broehan et al., 2010; Razaak et al., 2014; Wu et al., 2013; Panayides et al., 2015; Poletti et al., 2013). The IoT and video can be used for patients, particularly elderly patients being monitored from their home. The IoT collects data, such as temperature, blood rate and heart rate, using wearable technology. Video cameras in a room can be static or moving and provide real-time images and data for analysis. Smart cameras supported by IoT sensors can send an SMS or alarm in the case of any abnormal patient situation. Ahmed et al. (2021) proposed a monitoring system that uses deep learning in the classification of patient situations using IoT and video data. The proposed algorithm has achieved greater than 94% accuracy.

Virtual reality (VR) and game technologies are also commonly used in the health sector. They can be used in the training of doctors and nurses, especially for surgery training. Before performing a procedure, surgeons can practice it in a controlled, safe, and entirely digital environment using virtual reality. Physical therapy, relaxing patients, monitoring patients, disease awareness and medical marketing are other application areas of VR and games in the health sector (Yanhong et al., 2014; Shim et al., 2020; Zhou et al., 2021; Hassan et al., 2016; Li et al., 2018; Burdea et al., 2020). Using VR, doctors can undertake high-quality surgery training. A 360-degree angle from several cameras and an additional model create efficient surgery training for doctors and other health sector workers. This approach can be used in robotic surgery, patient consultation, mental health therapy and pain management. Serious games are computer-based games that can be used in the health sector to educate sector workers and for treatment purposes for some diseases, such as Parkinson's disease (Kusano et al., 2017; Torner et al., 2019; Gupta et al., 2019; Tsai et al., 2011; Cecil and Pirela-Cruz, 2013). A serious game helps patients significantly enhance their quality of life by allowing them to receive treatment and physical therapy from their own home.

AI technologies have several functionalities in the health sector, such as for clinics, patients and sector. Clinics can use technologies in decision-making (Jahan and Tripathi, 2021; Ladgham et al., 2013; Minz and Mahobiya, 2017), up-to-date information collection (Zhou et al., 2020; Mary et al., 2020; Rong et al., 2020) and information sharing (Wu et al., 2019; Gandhi et al., 2019; Zheng et al., 2018; Jaiman and Urovi, 2020).

In addition, machine learning improves the decision-making process by providing a solution for health-care practitioners and researchers to extract hidden knowledge from the huge amount of available data that cannot be discovered if relying on human effort only. A considerable amount of research has attempted to exploit machine learning to improve and automate the decision-making process in the health-care field (Thakkar et al., 2010; Moein et al., 2010; Nimmagadda et al., 2011; FengGuo et al., 2010; Bennett et al., 2011; Sqalli and Al-Thani, 2019; Minz and Mahobiya, 2017; Ribbens et al., 2019; Seeböck et al., 2019; Strachna and Asan, 2020; McGregor et al., 2020; Comito et al., 2020; Albu, 2017).

Patients are the main concern of health activities. AI technologies can be used for various applications from making appointments to patient monitoring. AI and other information technologies mainly have functionalities in patient diagnosis (Bernardini et al., 2021; Jahan and Tripathi, 2021; Ladgham et al., 2013; Patii and Iyer, 2017; Yang et al., 2018), treatment (Ciprian et al., 2020; Minz and Mahobiya, 2017; Zhou et al., 2021; Hassan et al., 2016), consultation (Sqalli and Al-Thani, 2019; Ladgham et al., 2013; Yu and Zhou, 2021), and health monitoring (Rahman et al., 2020; Maduri et al., 2020; Sakkos et al., 2021).

Many patients are in a healthy state where they need continuous monitoring. They can be either people who are in healthy state or where they cannot take care of themselves or senior patients who need continuous observation due to health or age-related concerns. This type of health monitoring consumes time, human resources and money, which is challenging but they cannot be avoided for patient safety and well-being when following traditional patient monitoring strategies. An innovative solution consists of remote patient monitoring. “Remote patient monitoring is an emerging field of healthcare, which concerns the management of health/illness with the goal of treating or diagnosing illness using information technology and telecommunication tools” (Jeddi and Bohr, 2020). Remote patient monitoring brings many advantages to both patients and hospitals. Tele-health systems are also getting important especially with the COVID-19 pandemic. Jiang (2022) developed a wearable tele-heath system to monitor especially patient and elderly people.

AI reduces hospital loads, resource consumption, hospital occupancy, and wasted time and effort in unnecessary medical intervention by minimizing the number of false alerts and relieving practitioners’ burdens to free their time for more important tasks. In addition, remote patient monitoring brings many benefits for patients, particularly elderly patients. First, it helps them avoid wasting time, effort and money by minimizing unnecessary visits to the hospital. Second, it plays a major role in safety and well-being by sending lifesaving alerts when the situation of the patient requires medical intervention. This is even more efficient when the patient is in weak health and cannot call for help. A great deal of research has been performed to implement and test health monitoring systems that combine biomedical data, collection devices and machine learning (Khan et al., 2020; Maduri et al., 2020; Chauhan et al., 2021; Ahmed et al., 2021). For example, Khan et al. (2020) encourage the inclusion of AI in medical practices to overcome the continuous increase in the number of patients. They state that including AI in medical practice enables doctors to better manage the non-stop biomedical information stream. The authors introduce a prototype that combines an ECG patch and machine learning to monitor patients. The EGC devices collect real-time biomedical data to check the health state of an individual's heart anywhere and anytime. The prototype processes the collected data using an algorithm that exploits multiple machine-learning techniques, including decision trees, regression trees, neural networks, deep learning and support vector machines. The output of these techniques is generated over ensemble methods. This prototype has been tested on adult patients at the Emory University Hospital (EUH). The authors claim that the model succeeded in eliminating unnecessary repetitive tasks, which provide more time for effective treatment tasks and reduce money and effort waste.

Multiple healthcare departments and sectors can use AI by following a multi-disciplinary approach in data collection using the IoT (Wu et al., 2019; Rahman et al., 2020; Maduri et al., 2020; Patii and Iyer, 2017), medical image processing (Jahan and Tripathi, 2021; Ladgham et al., 2013; Minz and Mahobiya, 2017), research development (Wu et al., 2013; Zhang et al., 2020; Azghadi et al., 2020) and remote surgery (Yanhong et al., 2014; Shim et al., 2020; Zhou et al., 2021; Hassan et al., 2016). A substantial quantity of research has attempted to further extend the medical field by both AI and the IoT. The latter is useful in connecting the infrastructure of the internet and sensor networks, which generate a useful stream of data that can be analyzed using AI models (Maduri et al., 2020; Chauhan et al., 2021; Ahmed et al., 2021).

Diagnosis is a crucial step to achieve successful treatment. Nevertheless, it is very challenging for many diseases, especially in their early stages. Nevertheless, early diagnosis can be a game changer for multiple diseases, as it can help patients, doctors and even hospitals countless effort and resources. For example, patients at the stage of melanomas of a thickness below 1 millimeter have a 95% 5-year survival rate. However, patients at the stage of melanomas of a thickness over 4 mm have only a 45% 5-year survival rate (Balch et al., 2001). The use of AI in measuring and detecting these physical conditions may prove to be the matter of life or death in these cases.

A considerable amount of research has investigated the potential of AI in early diagnosis (Gayathri et al., 2015; Katarya and Srinivas, 2020; Chauhan et al., 2021; Harinishree et al., 2021; Amrane et al., 2018; Charan et al., 2018). Cancer is one illness where early diagnosis may significantly alter the outcome. Early diagnosis of cancer can substantially increase the chance of survival by increasing the chances of treatment success. This can be even more beneficial when spotting cancer before it has the chance to spread.

Gayathri et al. (2015) suggested a model based on fuzzy logic that detects the risk of breast cancer. This model attempts to reduce the duration of diagnosis by implementing linear discriminant analysis (LDA) for feature reduction. The model performs training using a fuzzy inference method called the Mamdani fuzzy inference model. The authors claim that the system achieves an accuracy of 93%. Katarya and Srinivas (2020) approached the early diagnosis of heart diseases using AI. The authors performed a comparative study between AI algorithms using the existing Cleveland database. They found that decision tree and naïve Bayes outperform other models. The authors recommend using search algorithms for feature selection to achieve good accuracy. Murray et al. (2019) consider machine learning solutions to solve the semantic challenges facing health-care practitioners.

The findings outlined from the classification framework is illustrated in Fig. 6.

Fig. 6.

Mapping the classification framework across the patient, healthcare organization and healthcare sector

(0.25MB).