Content uploaded by Said El Kafhali
Author content
All content in this area was uploaded by Said El Kafhali on Jun 14, 2021
Content may be subject to copyright.
Artificial Intelligence for Healthcare:
Roles, Challenges, and Applications
Said El Kafhali and Mohamed Lazaar
Abstract With the use of an intelligent technology-based healthcare technique, there
may be a real opportunity to improve medical care quality and effectiveness, thereby
increasing patient wellness. Around the world, with rising healthcare costs and the
onset of many illnesses, it has become necessary to focus on the people-centered envi-
ronment, not just the hospital. The future of healthcare may change completely using
artificial intelligence (AI) that change how we prevent, diagnose, and cure health con-
ditions. However, the potential of AI is hard to ignore. It is a decision-making machine
that can exponentially increasethe efficiency ofthe healthcare organization. Recently,
many published papers use the AI technology to monitor and controls the spread of
COVID-19(Coronavirus) pandemic. There are not only the right set of circumstances
by using AI in healthcare but also many obstacles and barriers. Data integration is
complex, trust issues, time, and energy limitations are some of the barriers to imple-
menting AI in healthcare. Hence, this chapter provides a survey of AI-driven health-
care and identifies proposed models, which health staff is using to bring AI solutions
for health applications. It identifies existing approaches to designing models for AI
healthcare. The readers can benefit from the chapter by understanding the roles, chal-
lenges, applications, and future opportunities of AI for healthcare.
Keywords Artificial intelligence
·
Machine learning
·
Deep learning
·
Healthcare
1 Introduction
AI dates back to the 1950s as a university field and has been developing rapidly in
recent years with recent advances and innovations in information storage and pro-
cessing in which enabled intelligent systems using AI to revolutionize industries in
different fields [1]. The goal of AI is to develop technology that permits machines
S. El Kafhali (
B
)
Faculty of Sciences and Techniques, Computer, Networks, Mobility and Modeling Laboratory:
IR2M, Hassan First University of Settat, 26000 Settat, Morocco
e-mail: said.elkafhali@uhp.ac.ma
M. Lazaar
ENSIAS, Mohammed V University in Rabat, Rabat, Morocco
e-mail: mohamed.lazaar@um5.ac.ma
© The Author(s), under exclusive license to Springer Nature Switzerland AG 2021
N. Gherabi and J. Kacprzyk (eds.), Intelligent Systems in Big Data, Semantic Web and
Machine Learning, Advances in Intelligent Systems and Computing 1344,
https://doi.org/10.1007/978-3-030-72588-4_10
141
142 S. El Kafhali and M. Lazaar
and computers to operate intelligently. The key principle of AI is machine learning
(ML), or the capability of a machine to improve upon its skills by permanently ana-
lyzing its interactions with the real world. AI evenly refers to situations in which com-
puters can simulate human minds in learning and analysis and therefore can work on
problem-solving [2]. According to the types of problems that we want to solve, ele-
mentary ML algorithms can be divided into two classes: supervised machine learning
(SML) and unsupervised machine learning (USML) algorithms [3]. SML algorithms
work by collecting a big number of training cases that contain inputs and the required
output labels such as support vector machines, random forest, logistic regression,
most artificial neural networks, decision tree methods, fuzzy mathematical theory,
and so on. USML algorithms provide untagged data as the algorithms try to identify
the optimal parameters in the models to make sense by extracting functionality for the
learning case such as the K-means clustering, k nearest neighbors, self-organization
mapping model, principal component analysis, hierarchical cluster analysis, and so
forth.
Another approach used in AI is deep learning (DL), which is based on the con-
struction of artificial neural networks [4]. These networks, made up of thousands,
even millions of neurons, are inspired by the human brain. DL is often applied to
much larger amounts of data than ML and learns from this mass of data and in certain
cases obtains much better results than ML. It is particularly effective for working
with voice data. These voice data must be interpreted and translated into text before
a result can be found. The reinforcement learning (RL) model is an AI in which
machines identify the actions producing the highest probability of the result. It can
be formed by a set of trial and error sequences of events, exposing the model to
expert a combination of these strategies. This happens in a Markov decision process,
consisting of an ensemble of states, an ensemble of actions, the probability that some
action in some state will advance to a new state, and the reward that results from
the new state. Using the RL model, the machine establishes a policy that determines
the choice or action with the highest likelihood of the desired outcome, assessing
the total rewards attributable to the multiple actions performed over time and the
relative importance present and future rewards. However, AI habitually refers to a
machine that learns from raw data with some degree of autonomy, as occurs with
ML, DL, and RL as shown in Fig.1. With ML, DL, and RL, AI makes it possible
to solve problems that were thought to be reserved for human intelligence, such as
interpreting natural language or making predictions or complex recommendations.
AI has seen increasing use in all sectors, thanks to increasing data volumes, greater
computing power, and new algorithm architectures. In healthcare, there has been an
exponential increase in AI research, as evidenced by an increase in publications
and university funding. Most interestingly, in healthcare, the introduction of AI-
based technologies have reduced costs, improved analysis and treatment outcomes,
accelerated drug discovery and, consequently, increment the efficiency of the entire
healthcare business [5]. AI assists healthcare professionals, particularly to help them
identify the patients most at risk. It also plays an important role in preventing and
detecting diseases in real-time to save people’s lives. AI focuses on the analysis of
big data aggregated in healthcare settings to improve and develop clinical decision
Artificial Intelligence for Healthcare: Roles, Challenges, and Applications 143
Fig. 1 Summary of AI techniques
support systems or to evaluate medical data for both quality assurance and acces-
sibility of health services [6]. While AI can provide substantial improvements over
traditional healthcare for different tasks, many researchers and scientists remain
skeptical of their use when medical applications are involved. These skepticisms
arise because the AI theory has not yet provided complete solutions and many issues
remain unanswered. However, this chapter presents a review of research using AI
in healthcare, providing a critical analysis of the relative merit and potential pitfalls
of the technique as well as its prospects. To summarize, the goal thus is to outline
recent advancements in AI technologies and their healthcare applications, identify
the challenges for further improvement in healthcare AI models, and summarize the
ethical involvements and economic cost of AI in healthcare.
The rest of this chapter is organized as follows. Section 2provides a review of the
literature on AI in healthcare. The types of AI relevance to healthcare are presented
in Sect. 3. Section 4summarizes some diagnosis and treatment applications. Section
5presents some discussion and future trends.
2 A Review of the Literature on AI in Healthcare
AI plays a more important role in healthcare. It is currently under development or
implementation for use in many targeted health applications, including patient mon-
itoring, medical diagnostics, clinical decision support, and health systems learning
[7,8]. Many AI algorithms and software are under development or developed to sup-
port clinical decision-making and the development of public health policies. These
AI algorithms usually use computerized predictive analytics algorithms to organize,
filter, and search for models in large data sets from different sources and render a
probability analysis that healthcare providers can take on quick and informed deci-
sions [9]. To date, the majority of jurisdictions do not permit these AI algorithms to
be the final decision-maker in healthcare. Instead, they are mainly used as a screening
way or as a diagnostic aid. There are many papers published on the subject of applied
AI in healthcare. In this section, we summarize published important works related
to AI in healthcare.
144 S. El Kafhali and M. Lazaar
Wu et al. [10] presented a system that uses logistic regression to classify triple-
negative (TN) breast cancer on ultrasound images. Grayscale and color Doppler
images are used to calculate Ultrasound images characteristics. The authors analyzed
Ultrasonic and clinical data of140 surgicallyconfirmed cases for thediagnosis of TN
and non-TN (NTN) breast cancer. Diagnostic performance was measured by the area
under the receiver operating characteristic curve (ROC) with a sensitivity of 86.96%
and a specificity of 82.91% on the dataset used. Subasi et al. [11] proposed a hybrid
algorithm to detect epileptic seizures in Electroencephalogram (EEG) records using
Support Vector Machines (SVMs) and Genetic Algorithms (GAs) ML techniques.
The obtained results have shown that the proposed algorithm can achieve a classifi-
cation accuracy of up to 99.38% for the EEG datasets used and is a powerful model
for neuroscientists to detect epileptic seizures in EEG. Authors in [12] modeled the
diabetes diagnosis using a backpropagation neural network (BPNN) and probabilis-
tic neural network (PNN). A comparative analysis concluded that the PNN model is
better than the BPNN model in terms of accuracy. For the PNN model and by using
75% of training data and 25% of testing data, they obtained a diagnostic accuracy
of 97.9%. Based on the obtained results, the proposed model can effectively save
doctors time and enhance diagnosis compared with the traditional diagnosis process.
Alfaras et al. [13] proposed an automatic and fast electrocardiograph (ECG)
arrhythmia classifier based ML approach known as Echo State Networks. The
obtained results achieved a sensitivity of 92.7% and a positive predictive value of
86.1% for ventricular ectopic beats, using lead II, and a sensitivity of 95.7% and
positive predictive value of 75.1% when using the lead V1. A new statistical learn-
ing method, namely least absolute shrinkage and selection operator (LASSO), is
proposed in [14] for removing redundant and irrelevant features from the ECG big
data set. The features obtained from LASSO are trained with popular ML algorithms
such as multi-class one-against-all support vector machine, artificial neural networks,
and K-nearest neighbor (K-NN). The experimental results show that the proposed
method of LASSO witha K-NN classifier is effective with a significant improvement
in recognition accuracy of 99.1379% compared to some other existing techniques.
Abdeldayem et al. [15] proposed an approach that exploits the spectro-temporal
changes of the ECG signal to establish a personal recognition system using both
short-time Fourier transform (STFT) and generalized Morse wavelets (CWT). The
SFTF system achieved an average accuracy of 97.85%whereas the CWT achieves an
average accuracy of 97.5%, over the eight studied databases. Prashanth et al. [16] use
Support Vector Machine (SVM), Naïve Bayes, Boosted Trees, and Random Forests
classifiers to predict and detect early Parkinson’s disease (PD). The authors use impor-
tant biomarkers such as cerebrospinal fluid (CSF) measurements and dopaminergic
imaging markers from 183 healthy normal and 401 early PD subjects obtained from
the Parkinson’s Progression Markers Initiative (PPMI) database. The obtained results
showed that the SVM classifier gave the best performance (96.40% accuracy, 97.03%
sensitivity, 95.01% specificity). Shrivastava et al. [17] proposed a neural network to
predict PD with a feature selection technique. For the experiments, the authors use a
real-life dataset of 166 persons including both healthy controls and affected persons.
The experimental results showed that the Binary Bat Algorithm outperforms tradi-
Artificial Intelligence for Healthcare: Roles, Challenges, and Applications 145
tional Algorithms like Genetic Algorithm, Particle Swarm Optimization (PSO), and
Modified Cuckoo Search with an accuracy of 93.60%.
Authors in [18] used an ML-based system that uses the features in Computed
tomography (CT) images taken from the parenchyma around lung nodules to iden-
tify cancerous nodules. The results showed a sensitivity of 100% and a specificity of
96%. Yang et al. [19] presented an ML-based system to predict the grade of Glioblas-
toma using radiomic features of Multiparametric-magnetic resonance imaging. The
obtained results show that the grade of Glioblastoma could be predicted with an accu-
racy of 92% using the proposed approach. Guncar et al. [20] used ML algorithms
to diagnose hematological disorders based solely on laboratory results. In the first
method, they used all the blood tests available. In a second method, they used only a
limited set more habitually measured during patient intake. They obtained a predic-
tion accuracy of 88% and 86% for the two methods used respectively by considering
the list of the five most probable diseases and 59% and 57% by considering only the
most probable disease.
The comparative study of the related literature has been done in this section. The
important related AI methods applied for healthcare used in the literature review are
shown in Table 1.
3 Types of AI of Relevance to Healthcare
Artificial intelligence presents many techniques that can be applied in Industry 4.0.
Most of these techniques have a great impact on the services of the healthcare field. Its
performance in the Healthcare field depends on its parameters and the quality of data.
Figure 2illustrates the major ML Methods in Healthcare. MLis not only one method
with learning but also rather a scientific discipline of them that focuses on how com-
puters learn from data. ML is used to determine complex models, and extract knowl-
edge, presenting novel ideas to users. ML combines two major fields such as mathe-
matics and computer science. Many models of ML can be constructed from these two
scientific disciplines and massive data. We can classify the ML methods according to
their learning mode. In general, learning can be divided into many classes like super-
vised learning, unsupervised learning, reinforcement learning, Semi-Supervised learn-
ing, etc. In this section, we detailed the most used learning.
Supervised Learning: The supervised learning technique is the most popular mode
of learning used for predictive analysis when the dataset is labeled. In the prepro-
cessing phase, we define a couple (input, output, or target) for each input, this oper-
ation named labeling of the dataset. Supervised learning focuses on two classes
of problems, classification, and regression. Both classes share the same concept of
utilizing the training datasets to estimate the output parameters. Many algorithms
like Support Vector Machines, Naïve Bayes, Nearest Neighbor, Linear Regression,
Decision Trees, Neural Networks, etc. are categorized under SML. The main dif-
ference between the regression and classification is that the output parameter in the
146 S. El Kafhali and M. Lazaar
Table 1 Summary of some important papers of the AI techniques in healthcare
Paper Purpose Domain Technique Topic Best performance
Wu et al. [10] Evaluate the scope of ML together
with quantitative ultrasound image
features for triple-negative breast
cancer diagnosis
Radiology Logistic regression Breast cancer Sensitivity= 86.96%
Specificity = 82.91%
Subasi et al. [11] Establish a hybrid algorithm for
epileptic seizure detection to deter-
mine the optimum parameters of
SVMs for EEG data classification
Neurology ML Seizure detection Accuracy= 99.38%
Li et al. [12] Use BPNN and PNN models to the
diabetes diagnosis
Ophthalmology AI neural networks Diabetes diagnosis Accuracy= 97.9%
Alfaras et al. [13] Present a fast ECG arrhythmia clas-
sifier based on an ML approach
known as Echo State Networks
Cardiology ML Arrhythmia Sensitivity= 95.7%
Positive predictive= 86.1%
Patro et al. [14] Classify and identify the ECG data Cardiology Statistical learning Biometric recognition Accuracy= 99.1379%
Abdeldayem et al. [15] Establish a personal recognition sys-
tem using the deep convolutional
neural network and the spectro-
temporal changes of the ECG signal
Cardiology Deep Convolutional neural network Biometric recognition Accuracy= 97.85%
Prashanth et al. [16] Propose ML algorithms to give good
accuracy detection of Parkinson’s
Disease using multimodal features
Neurology ML Parkinson’s Disease Accuracy=96.40%
Sensitivity= 97.03%
Specificity = 95.01%
Shrivastava et al. [17] Predict PD with feature selection
technique
Neurology Neural network Parkinson’s Disease Accuracy=93.60%
Uthoff et al. [18] Use an ML-based system to dis-
tinguish malignant and benign lung
nodules
Radiology ML Lungs Sensitivity= 100%
Specificity = 96%
Yang et al. [19] Predict the grade of Glioblastoma Radiology ML Brain Accuracy= 92%
Guncar et al. [20] Diagnose hematological disorders Hematology ML Biological hematology Accuracy= 88%
Specificity = 96%
Artificial Intelligence for Healthcare: Roles, Challenges, and Applications 147
Fig. 2 Machine learning in healthcare
regression is numerical (or continuous) while that for classification is categorical (or
discrete). Classification is the process permits to find or discover many parameters of
the model, which assist to categorize the data into many classes or in discrete values.
The classification process is applied to the data where it can be divided into binary
or multiple discrete labels. In classification, data is labeled under different labels
according to some parameters given in input and then the labels are predicted for the
data. Regression is the process permits to find many parameters of the model to dis-
tingue the data into numerical continuous values. In regression, the model identifies
the distribution movement depending on the historical data.
Many supervised learning methods have been applied in the healthcare field. The
main objective is to improve the performance of the disease diagnosis system. In
this paragraph, we describe the interesting contributions based on classification (dis-
ease diagnosis) and prediction (drug efficacy). For the first type of applications, many
methods based on Neural networks have been used for the diagnosis of diabetes [21],
a hybrid approach based on Support Vector Machine and Multilayer Perceptron with
a Backpropagation (BP) algorithm is used in the diagnosis of chronic renal failure on
the Internet of Medical Things (IoMT) platform [22], others researchers proposed a
deep learning model that integrates S-Mask R-CNN and Inception-v3 in the ultra-
sound image-aided diagnosis of prostate cancer [23], a Naïve Bayes classification
model integrated with temporal association rules (TARs) for coronary heart disease
diagnosis [24]. For the second application, a modeling approach based on sequential
dependencies to improve the prediction of the drug efficacy using recurrent neural
networks [25], a hybrid approach for predicting drug efficacy for cancer treatment
based on comparative analysis of chemosensitivity and gene expression data. This
approach combines supervised and unsupervised learning algorithms [26], an inter-
esting method permits to Predict novel drugs for SARS-CoV-2 using ML from a >
10 million chemical space [27].
Unsupervised Learning: In unsupervised learning, the dataset is without labeled
responses i.e. no output variable to predict. In this mode, the model tries to cre-
ate clusters or groups of individuals based on their similar attributes, or naturally
148 S. El Kafhali and M. Lazaar
occurring trends, patterns, or relationships. This is a real challenge task to judge
the performance of unsupervised learning models. The popularly used algorithms
are K-Means, Expectation maximization, association rules, density-based algorithm,
etc. are the part of USML. Unsupervised learning problems are grouped into many
types, although two main problems are often used in literature: they are Clustering
and Dimensionality Reduction. Clustering is an important concept, it mainly deals
to find or create clusters in a collection of uncategorized data. Dimensionality reduc-
tion or feature extraction is the process permits to reduce the number of variables of
higher-dimensional data to lower-dimensional data.
The rest of this paragraph will briefly present related work about the two major
applications of unsupervised learning methods in healthcare; disease subtype discov-
ery and disease target discovery. For the first application, many researchers present
a description of the treatment of Alzheimer’s disease based on subtype-specific M1
allosteric agonists [28], others authors proposed an analytic framework permits to
find of complex disease subgroups, this framework was able to subtype schizophrenia
subjects into diverse subgroups with different prognosis and treatment response [29].
For the second application, the ML algorithms can be applied to segment images and
compute object and spatial-based data before analysis [30], an approach computes
the most relevant feature subset by taking advantage of feature selection and extrac-
tion techniques [31], the researchers discuss how to improve drug-target interaction
(DTI) prediction using deep-learning models and other ML algorithms [32], a random
forest model guided association of adverse drug reactions within Vitro target-based
pharmacology [33].
Reinforcement Learning: Reinforcement learning is a mode of learning based on
an agent that learns via interaction and feedback with the environment. The methods
based on reinforcement learning solve a task by acting and receiving rewards for an
environment. In general, computers take appropriate actions to increase the recom-
pense to generate the best decision and maximize the reward use of reinforcement
learning. Mostly used reinforcement learning algorithms are Q-learning, Temporal
difference, deep adversarial networks.
We describe some methods based on reinforcement learning in the health field in
particular on experimental design and drug design. Computer-aided drug design has
become a key source for information, rationalization, and inventiveness. In this field,
researchers have used many reinforcement algorithms. The authors provide numer-
ous recent examples of applications for drug design [34]. A reinforcement learning
(RL)-based optimal adaptive control approach is proposed for the continuous infu-
sion of a sedative drug to maintain a required level of sedation. In this study, integral
reinforcement learning (IRL) algorithm is designed to provide optimal drug dosing
for a givenperformance measure that iteratively updates the control solution concern-
ing the pharmacology of the patient while guaranteeing convergence to the optimal
solution [35]. Several machine-learning methods based on reinforcement learning
are used to predict drug design and discovery, which are efficient. An approach hybrid
a metaheuristic algorithm with support vector machine and k-Nearest Neighbors is
used for chemical descriptor selection and chemical compound activities [36].
Artificial Intelligence for Healthcare: Roles, Challenges, and Applications 149
Fig. 3 Machine learning in healthcare
4 Diagnosis and Treatment Applications
AI healthcare area has been a mostly upcoming hot topic in the present years. In recent
times, there has been an enormous thrust in the usage of AI for many applications,
particularly in the healthcare and pharmaceutical areas. Figure 3summarizes the
important applications of AI in healthcare.
Lifestyle: By using AI, we can understand our interactions with the environment, our
lifestyle improvements, and disease control with the ultimate purpose of improving
our well-being. Preventing disease is more important and effective than treating
disease. Therefore, if we can accurately predict our disease, we can control it properly
and effectively. For instance, the authors of [37] developed prediction models based
on AI to control type-2 diabetes through guided lifestyle management.
Digital Health Monitoring and Diagnostics: In the past, healthcare decision-
making and diagnosis was solely based on the physician’s personal experience,
health knowledge, the patient’s physical signs and symptoms, and laboratory diag-
nostic analysis. Today, in a typical healthcare monitoring system, Medical Internet
of Things (MIoT) devices have emerged to play an important role in helping physi-
cians and medical personnel to obtain accurate data and help reach a more accurate
assessment, diagnosis, and decision [38]. These MIoT are designed to collect health
data from the patient, which needs to be stored and analyzed or even researched in
the future [39]. To analyze and take the real-time decision from the big health data
collected by IoT, AI methods are needed to save patient life especially in the case of
cardiac health monitoring [40].
Surgery: AI has the potential to change the manner surgery is practiced with the
engagement of an optimized future for the highest quality patient care and surgeon
workflow [41]. Even though robotic autonomous surgery will stay out of reach for
some time, the synergy between the domains can probably expedite the competences
of AI to raise surgical care. Therefore, surgeons should be engaged in the assess-
ment of the quality and applicability of AI progress to ensure proper translation in
the clinical sector. For example, AI can enhance plastic surgery practice and help
plastic surgeons with the skills, knowledge, and tools to improve patient care in the
future [42]. Integrating AI into surgical decision-making has the potential to change
150 S. El Kafhali and M. Lazaar
care by increasing the decision to operate, recognition and attenuation of modifiable
risk factors, informed consent process, decisions about postoperative management,
and shared decisions regarding the use of resources [43].
Mental Health: The future of AI in mental health is hopeful. AI can have an impact
on psychological and psychiatric care in terms of information gathering, diagnosis,
and treatment. Recently, mental health exercise yet has much to benefit from AI tech-
nology. Therefore, authors in [44] reviewed numerous papers that used AI-based on
electronic health records, brain imaging data, mood scales, new surveillance systems,
and social media platforms to predict, classify or sub-group mental health illnesses,
including depression [45], schizophrenia [46] or other psychiatric illnesses [47], and
suicide attempts and ideas [48].
Virtual Assistant: In eHealth, the use of virtual assistant (VA) is effective and
acceptable to older adults, including those with limited health knowledge. It should
be able to establish a relationship with the patient over a series of interactions.
The VA presents an intelligent behavior similar to a human in the way each daily
dialogue takes place and ensuring the consistency of the daily interventions towards
the desirable behaviors [49]. With the application of AI, surgical robots, the diagnosis,
and treatment of diseases have become smarter such as the use of VA to improve the
mental health of humans. The AI accuracy diagnosis results exceed that of human
doctors [50]. Overall, VA mainly acts as a bridge to communicate with doctors,
patients, and medical staff. For patients, the VA can easily convert daily language
into medical language thanks to the smart device, to more precisely search for the
corresponding medical services. For doctors, the VA can automatically respond to
relevant information based on basic patient information, helping doctors to manage
patients and coordinate medical procedures more easily, so doctors can save more
time. For medical staff, the application of VA can significantly save human and
material resources and meet the needs of all parties more effectively.
Hospital Management: AI in healthcare is changing the obverse of hospital man-
agement [51]. The help of AI in healthcare will improve the performance of hospitals,
doctors, and nurses and provide patients with more targeted and personalized ser-
vices. AI makes healthcare facilities more efficient and improves the lives of health-
care providers and patients by automating tasks in the least amount of time and
money. Numerous medical tasks made easier by AI in healthcare such as improving
medical records and treatment solutions, automating customer relationship manage-
ment, monitoring vital statistics of ICU (Intensive Care Units) patients, simplified
health insurance verification, decoding of laboratory results, greater efficiency in
hospital operations, facilitating patient engagement and adherence, enhancing the
administrative activities and so forth.
Patient Data and Risk Analysis: AI can help us to analyze the patient data in
real-time to save a life. For example, recently AI is one of the news technologies
that can easily track the spread of COVID-19 virus, identify high-risk patients, and
is useful in controlling this infection in real-time. It can also predict the risk of
mortality by adequately analyzing past patient data [52]. AI technologies have also
Artificial Intelligence for Healthcare: Roles, Challenges, and Applications 151
been widely used to analyze clinical data, including electronic health records, medical
images, and physiological signals [53]. AI methodologies have been adopted also
to extract information from the data collected at various stages of drug development
that contains information about the mechanisms and treatments of the disease.
Wearables: Wearable devices facilitate the development of algorithms for the
automated prediction, prevention and response of health events in many areas:
metabolic, cardiovascular, gastrointestinal monitoring, sleep, neurology, movement
disorders, mental health, maternal, pulmonary health, and environmental exposures
and so forth [54]. Wearables also provide ongoing medical data to actively monitor
metabolic status, diagnosis, and treatment [55]. Whereas progress in the development
of wearable and AI will undoubtedly increase over the next few decades, random-
ized clinical studies are required to assess their real impact on patient care. In the
future, the wearable will not only be diagnostic, preventive, and therapeutic methods
but will also allow the uninterrupted acquisition of health data to monitor disease
progression, response to drugs, and assess the effectiveness of clinical trials.
Drug Discovery, Research, and Development: At present, exceptional improve-
ments in computing power married with advances in AI technology, could be used to
revolutionize the drug discovery and development process [56]. The drug develop-
ment process follows an inductive-deductive cycle, which ultimately leads to opti-
mized hit and leads compounds. The first step is the identification of new chemical
compounds with biological activity. The biological activity can result from the inter-
action of the compound with a specific enzyme or with an entire organism [57].
The second step is the identification of a lead molecule, which is a chemical com-
pound that can lead to the development of a new drug to treat a disease. Using AI
to automate the identification of new chemical compounds or the identification of a
lead molecule can reduce errors and improve the efficiency of drug discovery and
development [58]. There are other uses of AI in drug discovery and development,
including pharmacological properties [59], predicting workable synthetic routes for
drug-like molecules [60], drug activity against cancer cells [61], protein structural
features [62], drug combination [63], drug research [64], drug reuse [65] and many
others [66]. Besides, the identification of new pathways and targets using omics
analysis becomes possible through the generation of new biomarkers and therapeu-
tic targets, personalized medicine basedon omics markers, and the discovery of links
between drugs and diseases [67].
5 Conclusions and Future Trends
Most of the past studies applied the AI algorithms on healthcare are either focusing
on unsupervised or supervised methods. Unsupervised techniques use clustering or
more complex means to identify structure in data to present the data visualizations or
summaries. Supervised techniques, on the other hand, focus on finding very specific
and predefined type models or constructing predictive models. No matter which of
152 S. El Kafhali and M. Lazaar
these techniques is used, the families of underlying models or the similarity metric
push a strong bias in the analytical process. Therefore, current AI algorithms on
healthcare applications mainly focus on answering rather well-asked questions. One
could support that this type of healthcare problem-solving was suitable before when
healthcare data resources were significantly little and could be hoped to make sense
of it using such limited techniques. To date, health care data has greatly exceeded
our capacity to analyze it, and new powerful and flexible AI techniques are needed
that allow the unexpected to be discovered, allowing medical personnel and the
physician users to formulate new hypotheses interactively and help them discover
real perspectives of understanding of the disease. Therefore, we need to develop
discovery support AI algorithms rather than automated discovery AI algorithms.
Such an AI algorithm should not try to do the discovery work for medical personnel
and the physician users- it should rather support them by giving associative and
intuitive access to everything the AI algorithm has access to unstructured and semi-
structured data up to elements of expert knowledge humanly annotated.
There are various problems with ML algorithms to take into consideration when
we apply these clinically [68]. The first one is its issue with overfitting which is a
real issue along with every prediction method [69]. It will perform exceptionally too
well on its training data, but very poorly on new data. This means that the false signal
or noise in the training dataset is picked up and learned as concepts by the model.
Overfitting takes place when the training algorithm acknowledges the false signal
or noise in the dataset as a signal and carries out the prediction to the test dataset,
showing in bad performance on the new health data or an incapability to validate the
model externally. There are divers’ methods to solve and recognize the problem of
overfitting. First, if the precision changes drastically, for example, the 99% precision
on the training set drops to 50% when the algorithm is applied to the new dataset.
Second, if the precision on the training dataset increases, but the precision on the
validation dataset stays the same or decreases, we need to stop training. Thirdly, the
goodness of fit test can measure how well the predicted values of the algorithm match
the observed true values. Finally, when we have various comparable algorithms, we
can use the simpler ones so that the added benefit of any complexity can be given,
which favors a simpler model among others. Besides, the issue of overfitting can be
further reduced by a sampling technique including cross-validation that repeatedly
partitions the sample data without method or conscious decision into training and
validation sets to validate model predictions internally.
One more issue is that most clinicians and perhaps some researchers are unaware
of what an ML algorithm does to produce its output. A popular example is a project
to explore the outcomes of pneumonia-related hospitalization in the 1990s [70], in
which asthma has appeared as a protective element against pneumonia in the project.
For that, the clinicians should know from their experience that co-morbid asthma
is not a protective element. In a complex machining operation generally, process
parameters bear very complex mathematical relationships among them. However, as
the process is costly it is not possible to generate a large dataset. In this case, we
Artificial Intelligence for Healthcare: Roles, Challenges, and Applications 153
can use ML techniques with such a low number of data for supervised learning to
predict and optimize the machining process parameters, as often ML employed for
the problem with a large dataset.
References
1. Garbuio, M., Lin, N.: Artificial intelligence as a growth engine for health care startups: emerg-
ing business models. Calif. Manage. Rev. 61(2), 59–83 (2019)
2. Rong, G., Mendez, A., Assi, E.B., Zhao, B., Sawan, M.: Artificial intelligence in healthcare:
review and prediction case studies. Engineering 6(3), 291–301 (2020)
3. Chang, Z., Du, Z., Zhang, F., Huang, F., Chen, J., Li, W., Guo, Z.: Landslide susceptibility pre-
diction based on remote sensing images and GIS: comparisons of supervised and unsupervised
machine learning models. Remote Sensing 12(3), 502 (2020)
4. Ravì, D., Wong, C., Deligianni, F., Berthelot, M., Andreu-Perez, J., Lo, B., Yang, G.Z.: Deep
learning for health informatics. IEEE J. Biomed. Health Inform. 21(1), 4–21 (2016)
5. Yu, K.H., Beam, A.L., Kohane, I.S.: Artificial intelligence in healthcare. Nat. Biomed. Eng.
2(10), 719–731 (2018)
6. Lysaght, T., Lim, H.Y., Xafis, V., Ngiam, K.Y.: AI-assisted decision-making in healthcare.
Asian Bioethics Rev. 11(3), 299–314 (2019)
7. Reddy, S., Allan, S., Coghlan, S., Cooper, P.: A governance model for the application of AI
in health care. J. Am. Med. Inform. Assoc. 27(3), 491–497 (2020)
8. Magrabi, F., Ammenwerth, E., McNair, J.B., De Keizer, N.F., Hyppönen, H., Nykönen, P.,
Rigby, M., Scott, P.J., Vehko, T., Wong, Z.S.Y., Georgiou, A.:Artificial Intelligence in clinical
decision support: challenges for evaluating AI and practical implications. Yearbook Med.
Informa. 28(01), 128–134 (2019)
9. Persson, A., Kavathatzopoulos, I.: How to make decisions with algorithms: ethical decision-
making using algorithms within predictive analytics. ACM SIGCAS Comput. Soc. 47(4),
122–133 (2018)
10. Wu, T., Sultan, L.R., Tian, J., Cary, T.W., Sehgal, C.M.: Machine learning for diagnostic
ultrasound of triple-negative breast cancer. Breast Cancer Res. Treat. 173(2), 365–373 (2019)
11. Subasi, A., Kevric, J., Canbaz, M.A.: Epileptic seizure detection using hybrid machine learning
methods. Neural Comput. Appl. 31(1), 317–325 (2019)
12. Li, X.: Artificial intelligence neural network based on intelligent diagnosis. J. Ambient Intell.
Hum. Comput. (2020). https://doi.org/10.1007/s12652-020-02108-6
13. Alfaras, M., Soriano, M.C., Ortín, S.: A fast machine learning model for ECG-based heartbeat
classification and arrhythmia detection. Front. Phys. 7, 103 (2019)
14. Patro, K.K., Reddi, S.P.R., Khalelulla, S.E., Kumar, P.R., Shankar, K.: ECG data optimization
for biometric human recognition using statistical distributed machine learning algorithm. J.
Supercomput. 76(2), 858–875 (2020)
15. Abdeldayem, S.S., Bourlai, T.: ECG-based human authentication using high-level spectro-
temporal signal features. In: 2018 IEEE International Conference on Big Data (Big Data), pp.
4984–4993. IEEE Press (2018)
16. Prashanth, R., Roy, S.D., Mandal, P.K., Ghosh, S.: High-accuracy detection of early Parkin-
son’s disease through multimodal features and machine learning. Int. J. Med. Inform. 90,
13–21 (2016)
17. Shrivastava, P., Shukla, A., Vepakomma, P., Bhansali, N., Verma, K.: A survey of nature-
inspired algorithms for feature selection to identify Parkinson’s disease. Comput. Methods
Prog. Biomed. 139, 171–179 (2017)
154 S. El Kafhali and M. Lazaar
18. Uthoff, J., Stephens, M.J., Newell Jr., J.D., Hoffman, E.A., Larson, J., Koehn, N., De Ste-
fano, F.A., Lusk, C.M., Wenzlaff, A.S., Watza, D., Neslund-Dudas, C.: Machine learning
approach for distinguishing malignant and benign lung nodules utilizing standardized perin-
odular parenchymal features from CT. Med. Phys. 46(7), 3207–3216 (2019)
19. Cui, G., Jeong, J.J., Lei, Y., Wang, T., Liu, T., Curran, W.J., Mao, H., Yang, X.: Machine-
learning-based classification of Glioblastoma using MRI-based radiomic features. In: Pro-
ceedings of SPIE 10950, Medical Imaging 2019: Computer-Aided Diagnosis, 1095048, March
2019. https://doi.org/10.1117/12.2513110
20. Gunˇcar, G., Kukar, M., Notar, M., Brvar, M., ˇ
Cernelˇc, P., Notar, M., Notar, M.: An application
of machine learning to haematological diagnosis. Sci. Rep. 8(1), 1–12 (2018)
21. Erkaymaz, O., Ozer, M., Perc, M.: Performance of small-world feedforward neural networks
for the diagnosis of diabetes. Appl. Math. Comput. 311, 22–28 (2017)
22. Ma, F., Sun, T., Liu, L., Jing, H.: Detection and diagnosis of chronic kidney disease using
deep learning-based heterogeneous modified artificial neural network. Fut. Gener. Comput.
Syst. 111, 17–26 (2020)
23. Liu, Z., Yang, C., Huang, J., Liu, S., Zhuo, Y., Lu, X.: Deep learning framework based on
integration of S-Mask R-CNN and Inception-v3 for ultrasound image-aided diagnosis of
prostate cancer. Fut. Gener. Comput. Syst. 114, 358–367 (2020)
24. Orphanou, K., Dagliati, A., Sacchi, L., Stassopoulou, A., Keravnou, E., Bellazzi, R.: Incor-
porating repeating temporal association rules in Naïve Bayes classifiers for coronary heart
disease diagnosis. J. Biomed. Inform. 81, 74–82 (2018)
25. Kang, S.: Personalized prediction of drug efficacy for diabetes treatment via patient-level
sequential modeling with neural networks. Artif. Intell. Med. 85, 1–6 (2018)
26. Wan, P., Li, Q., Larsen, J.E.P., Eklund, A.C., Parlesak, A., Rigina, O., Nielsen, S.J., Bjöorkling,
F., Jónsdóttir, S.Ó.: Prediction of drug efficacy for cancer treatment based on comparative
analysis of chemosensitivity and gene expression data. Bioorg. Med. Chem. 20(1), 167–176
(2012)
27. Kowalewski, J., Ray, A.: Predicting novel drugs for SARS-CoV-2 using machine learning
from a 10 million chemical space. Heliyon 6(8), e04639 (2020)
28. Lebois, E.P., et al.: Towards the treatment of Alzheimer’s disease: discovery and development
of novel subtype-specific M1 allosteric agonists. Alzheimers Dement. 5(4), P335 (2009)
29. Yin, L., Chau, C.K., Sham, P.C., So, H.C.: Integrating clinical data and imputed transcriptome
from GWAS to uncover complex disease subtypes: applications in psychiatry and cardiology.
Am. J. Hum. Genet. 105(6), 1193–1212 (2019)
30. Caie, P.D., Dimitriou, N., Arandjelovi´c, O.: Precision medicine indigital pathology via image
analysis and machine learning. In: Artificial Intelligence and Deep Learning in Pathology, pp.
149–173. Elsevier (2021)
31. Shah, S.M.S., Shah, F.A., Hussain, S.A., Batool, S.: Support vector machines-based heart
disease diagnosis using feature subset, wrapping selection and extraction methods. Comput.
Electr. Eng. 84, 106628 (2020)
32. D’Souza, S., Prema, K.V., Balaji, S.: Machine learning models for drug-target interactions:
current knowledge and future directions. Drug Disc. Today 25(4), 748–756 (2020)
33. Ietswaart, R., Arat, S., Chen, A.X., Farahmand, S., Kim, B., DuMouchel, W., Armstrong, D.,
Fekete, A., Sutherland, J.J., Urban, L.: Machine learning guided association of adverse drug
reactions with in vitro target-based pharmacology 57, 102837 (2020)
34. Daina, A., Röhrig, U.F., Zoete, V.: Computer-aided drug design for cancer therapy. In: Wolken-
hauer, O. (ed.) Systems Medicine, pp. 386–401. Academic Press, Oxford (2021)
35. Padmanabhan, R., Meskin, N., Haddad, W.M.: Optimal adaptive control of drug dosing using
integral reinforcement learning. Math. Biosci. 309, 131–142 (2019)
36. Houssein, E.H., Hosney, M.E., Oliva, D., Mohamed, W.M., Hassaballah, M.: A novel hybrid
Harris hawks optimization and support vector machines for drug design and discovery. Com-
put. Chem. Eng. 133, 106656 (2020)
37. Hsu, G.C.: Using math-physical medicine and artificial intelligence technology to manage
lifestyle and control metabolic conditions of T2D. Int. J. Diab. Complications 2(3), 1–7
(2018)
Artificial Intelligence for Healthcare: Roles, Challenges, and Applications 155
38. El Kafhali, S., Salah, K.: Performance modelling and analysis of Internet of Things enabled
healthcare monitoring systems. IET Netw. 8(1), 48–58 (2018)
39. El Kafhali, S., Salah, K., Alla, S.B.: Performance evaluation of IoT-fog-cloud deployment for
healthcare services. In: 2018 4th International Conference on Cloud Computing Technologies
and Applications (Cloudtech), pp. 1–6. IEEE, November 2018
40. Zhang, Q.: Artificial intelligence-enabled ECG big data mining for pervasive heart health
monitoring. In: Biomedical Signal Processing, pp. 273–290. Springer, Singapore (2020)
41. Hashimoto, D.A., Rosman, G., Rus, D., Meireles, O.R.: Artificial intelligence in surgery:
promises and perils. Ann. surg. 268(1), 70 (2018)
42. Liang, X., Yang, X., Yin, S., Malay, S., Chung, K.C., Ma, J., Wang, K.: Artificial intelligence
in plastic surgery: applications and challenges. Aesth. Plast Surg. 1–7 (2020)
43. Loftus, T.J., Tighe, P.J., Filiberto, A.C., Efron, P.A., Brakenridge, S.C., Mohr, A.M., Rashidi,
P., Upchurch, G.R., Bihorac, A.: Artificial intelligence and surgical decision-making. JAMA
Surg. 155(2), 148–158 (2020)
44. Graham, S., Depp, C., Lee, E.E., Nebeker, C., Tu, X., Kim, H.C., Jeste, D.V.: Artificial
intelligence for mental health and mental illnesses: an overview. Curr. Psychiatry Rep. 21(11),
116 (2019)
45. Anis, K., Zakia, H., Mohamed, D., Jeffrey, C.: Detecting depression severity by interpretable
representations of motion dynamics. In: 2018 13th IEEE International Conference on Auto-
matic Face & Gesture Recognition (FG 2018), pp. 739–745. IEEE (2018)
46. Kalmady, S.V., Greiner, R., Agrawal, R., Shivakumar, V., Narayanaswamy, J.C., Brown,
M.R., Greenshaw, A.J., Dursun, S.M., Venkatasubramanian, G.: Towards artificial intelli-
gence in mental health by improving schizophrenia prediction with multiple brain parcellation
ensemble-learning. npj Schizophrenia 5(1), 1–11 (2019)
47. Thakur, A., Alam, M.S., Abir, M.R.H., Kushal, M.A.A., Rahman, R.M.: A fuzzy approachfor
the diagnosis of depression. In: Modern Approaches for Intelligent Information and Database
Systems, pp. 199–211. Springer, Cham (2018)
48. Cook, B.L., Progovac, A.M., Chen, P., Mullin, B., Hou, S., Baca-Garcia, E.: Novel use of
natural language processing (NLP) to predict suicidal ideation and psychiatric symptoms in
a text-based mental health intervention in Madrid. Comput. Math. Methods Med. (2016)
49. Balsa, J., Neves, P., Félix, I., Guerreiro, M.P., Alves, P., Carmo, M.B., Marques, D., Dias, A.,
Henriques, A., Cláudio, A.P.: Intelligent virtual assistant for promoting behaviour change in
older people with T2D. In: EPIA Conference on Artificial Intelligence, pp. 372–383. Springer,
Cham (2019)
50. Tian, S., Yang, W., Le Grange, J.M., Wang, P., Huang, W., Ye, Z.: Smart healthcare: making
medical care more intelligent. Global Health J. 3(3), 62–65 (2019)
51. Gu, D., Li, J., Bichindaritz, I., Deng, S., Liang, C.: The mechanism of i nfluence o f a case-based
health knowledge system on hospital management systems. In: International Conference on
Case-Based Reasoning, pp. 139–153. Springer, Cham (2017)
52. Vaishya, R., Javaid, M., Khan, I.H., Haleem, A.: Artificial Intelligence (AI) applications for
COVID-19 pandemic. Diabet. Metab. Syndr.: Clin. Res. Revi. 14(4), 337–339 (2020)
53. Wang, F., Preininger, A.: AI in health: state of the art, challenges, and future directions.
Yearbook Med. Inform. 28(1), 16–26 (2019)
54. Dunn, J., Runge, R., Snyder, M.: Wearables and the medical revolution. Personalized Med.
15(5), 429–448 (2018)
55. Yetisen, A.K., Martinez-Hurtado, J.L., Ünal, B., Khademhosseini, A., Butt, H.: Wearables in
medicine. Adv. Mater. 30(33), 1706910 (2018)
56. Díaz, Ó., Dalton, J.A., Giraldo, J.: Artificial intelligence: a novel approach for drug discovery.
Trends Pharmacol. Sci. 40(8), 550–551 (2019)
57. Correia, J., Resende, T., Baptista, D., Rocha, M.: Artificial intelligence in biological activity
prediction. In: International Conference on Practical Applications of Computational Biology
& Bioinformatics, pp. 164–172. Springer, Cham (2019)
58. Mak, K.K., Pichika, M.R.: Artificial intelligence in drug development: present status and
future prospects. Drug Disc. Today 24(3), 773–780 (2019)
156 S. El Kafhali and M. Lazaar
59. Klopman, G., Chakravarti, S.K., Zhu, H., Ivanov, J.M., Saiakhov, R.D.: ESP: a method to pre-
dict toxicity and pharmacological properties of chemicals using multiple MCASE databases.
J. Chem. Inf. Comput. Sci. 44(2), 704–715 (2004)
60. Merk, D., Friedrich, L., Grisoni, F., Schneider, G.: De novo design of bioactive small molecules
by artificial intelligence. Mol. Inform. 37(1–2), 1700153 (2018)
61. Lind, A.P., Anderson, P.C.: Predicting drug activity against cancer cells by random forest
models based on minimal genomic information and chemical properties. PloS One 14(7),
e0219774 (2019)
62. Klausen, M.S., Jespersen, M.C., Nielsen, H., Jensen, K.K., Jurtz, V.I., Sønderby, C.K., Som-
mer, M.O.A., Winther, O., Nielsen, M., Petersen, B., Marcatili, P.: NetSurfP-2.0: improved
prediction of protein structural features by integrated deep learning. Proteins: Struct. Funct.
Bioinform. 87(6), 520–527 (2019)
63. Tsigelny, I.F.: Artificial intelligence in drug combination therapy. Briefings Bioinform. 20(4),
1434–1448 (2019)
64. Tu, H., Lin, Z., Lee, K.: Automation with intelligence in drug research. Clin. Ther. 41(11),
2436–2444 (2019)
65. Hui, T.K., Mohammed,B., Donyai, P., McCrindle, R., Sherratt, R.S.: Enhancing pharmaceuti-
cal packaging through a technology ecosystem to facilitate the reuse of medicines and reduce
medicinal waste. Pharmacy 8(2), 58 (2020)
66. Chan, H.S., Shan, H., Dahoun, T.,Vogel, H., Yuan, S.: Advancing drug discovery viaartificial
intelligence. Trends Pharmacol. Sci. 40(8), 592–604 (2019)
67. Villalta, F., Rachakonda, G.: Advances in preclinical approaches to Chagas disease drug
discovery. Exp. Opinion Drug Disc. 14(11), 1161–1174 (2019)
68. Lee, S., Mohr, N.M., Street, W.N., Nadkarni, P.: Machine learning in relation to emergency
medicine clinical and operational scenarios: an overview. West. J. Emerg. Med. 20(2), 219–
227 (2019)
69. Hawkins, D.M.: The problem of overfitting. J. Chem. Inf. Comput. Sci. 44(1), 1–12 (2004)
70. Craven, D.E., Steger, K.A., Barber, T.W.: Preventing nosocomial pneumonia: state of the art
and perspectives for the 1990s. Am. J. Med. 91(3), S44–S53 (1991)
