the fraud related to medical insurance will minimized to the least.
ML can transform the healthcare services by making us better providers of correct medical facility at the patient level. We can gather information on how different environmental exposure and lifestyle will vary the symptoms of disease. The intervention and history will help us decide treatment and decision-making. We can further understand the health and disease trajectory which will help in prepare us before arrival of the pandemics in worst possible situation. The resources available to us can be utilized in more efficient way with reduced costs. Also, the public health policies can be transformed in a way benefiting the people.
Figure 1.2 Application of ML in healthcare.
1.3 Machine Learning Algorithms
Depending upon the problem and approach to be applied, it has been categorized into various types among which major application lie into supervised, unsupervised, semi-supervised, reinforcement, and DL. The various types and its contribution to healthcare sector are shown in Figure 1.3.
1.3.1 Supervised Learning
This ML algorithm works under supervision, i.e., machine is trained with data which is well labeled and helps the model to predict with the help of dataset. Furthermore, supervised learning is divided into classification and regression. When the resultant variable is categorical, i.e., with two or more classes (yes/no, true/false, disease/no disease), we make use of classification. Whereas, when the resultant variable is a real and uninterrupted value, the problem is regression, here, a change in one variable is linked with a change in other variable (e.g., weight based on height). Some common examples of supervised ML in medicine is to perform pattern recognition over selected set of diagnosis by a cardiologist by interpretation of EKG and also from a chest X ray detection of lung module can be determined [5]. In medicine, for detection of risk in coronary heart disease, the best possible method adopted for analysis is Framingham Risk Score which is an example of supervised ML [6]. Risk models like above in medicine can guide in antithrombotic therapy in atrial fibrillation [7] and in hypertrophic cardiomyopathy for the implantation of electronic defibrillators [8].
Figure 1.3 The types of machine learning algorithm.
1.3.2 Unsupervised Learning
Such type of ML algorithm does not work upon labeled data and the machine learns from the dataset given and finds out the hidden pattern to make prediction about the output. It is further grouped into clustering and association; in clustering, the machine forms groups based on the behavior of the data, whereas association is a rule-based ML to discover relation between variables of large datasets. Precision medicine initiative is used to perform unsupervised learning problems in medicine [9]. How unsupervised learning can be applied in pathophysiologic mechanism to redefine the inherent heterogeneity in complex multi-factorial diseases, for instance, in cardiac disease like myocarditis. To apply the mechanism, inexplicable acute systolic heart failure is required and performed with myocardial biopsies to identify similar pattern between cellular compositions which will, in return, guide the therapist accordingly. Albiet the same technique to identify a subtype of asthma which responded to IL-13 [10, 11] is adopted.
1.3.3 Semi-Supervised Learning
It is a combination of supervised and unsupervised learning. It uses a combination of small portion of labeled data and massive collection of unlabeled to improve the prediction. The algorithm has ability to learn how to react on a particular situation based on the environment. The main aim of this method is to improve classification performance. This method is highly applicable in the healthcare sector when labeled data is not sufficiently available. It is applicable for classification of protein sequence typically due to the large size of DNA strands. Consistency enforcing strategy is mostly followed by this method [12]. It has been widely used for classification of medical images to reduce effort over labeling data [13–15]. Apart from this, for breast cancer analysis [16] and liver segmentation [17], a co-training mechanism has been applied.
1.3.4 Reinforcement Learning
This category of algorithm has no predefined data and the input depends upon the action taken by the agent and these actions are then recorded in the form of matrices so that it can serve as the memory to the agent. The agent explores the environment and collects data which is further used to get the output. In medicine, there are several instances of reinforcement learning (RL) application like for the development of therapy plan for lung cancer [18] and epilepsy [19]. Deep RL approach has been recently proposed for the therapy plan development on medical registry data [20] and also to learn treatment strategies for sepsis [21].
1.3.5 Deep Learning
Such algorithms has been widely used in the field of science for solving and analyzing problems related to healthcare by using different techniques for image analysis for obtaining information effectively. DL requires data to get information but, when combined with the medicinal data, makes the work complex for the researcher. Once the data is obtained, it can be applied accordingly in different field of medicine like prognosis, diagnosis, treatment, and clinical workflow. DL concept is used to build tool for skin cancer detection in dermatology [22]. Neural network training using DL method is applied for the computation of diabetic retinopathy severity by using the strength of pixels in fundus image [23].
1.4 Big Data in Healthcare
Sebastian Thrum, a computer scientist, once said, “Just as machines made human muscles a thousand times stronger, machines will make the human brain a thousand times more powerful.” This statement is the seed of what big data and ML is doing to healthcare today impacting human lives like never before opening new doors of possibilities and for much good.
In this digital era, massive amounts of data are being generated every moment, the digital universe which was about 130 exabytes (EB) in 2005 has expanded to about 40,000 EB in 2020 [24]. Such huge amount of data, known as big data, is a storehouse of critical information which can transform the way we provide healthcare services.
Figure 1.4 Sources of big data in healthcare.
Large amounts of data are generated in healthcare services, and these are from sources (Figure 1.4) as diverse as government agencies, patient portals, research studies, generic databases, electronic health records, public health records, wearable devices, smart phones, and many more. All these sources generate data in different formats which need to be not just merged but also made available instantly when needed; this is where big data and ML together empower healthcare services.
1.5
