3.1.3. Disease Tracking

AI can be deployed for tracing and tracking COVID-19. Current studies have shown that COVID-19 is characterized by respiratory patterns, which differ from normal colds and periodic influenza, including fast breathing (tachypnea) [82]. Predicting tachypnea could become a premium diagnostic characteristic that aids the scope screening of possible patients [55]. So many proposals have been made on how best to employ mobile phones for COVID-19 diagnoses. The best way is either by using embedded sensors that detect COVID-19 symptoms or by conducting phone surveys to help vulnerable patients who depend on responding to critical questions [83].

Berlin uses a model based on epidemiological SIR, which uses curtailment actions by the relevant authorities, such as quarantines, social distance, and partial or total lockdown measures [84]. Another SIR model involves public health methods for handling the virus. It also uses data sources from China and was made available in R [85]. GLEAMviz epidemiological model could be deployed to check the spread of the virus [33]. Similarly, Metabiota [86] uses a tracker for the epidemic to detect COVID-19 [87]. It is also used as a near-tenure forecasting system for the transmission of 93infections. Information about tracking the virus is essential for public health experts to curtail the pandemic [88,89] effectively.

#### 3.1.4. Prediction of the Infected Patient

An innovative method that depends on patients' blood tests and medical information was developed to assist doctors in determining vulnerable patients early enough. This will improve virus forecasting and reduce the mortality rate among high-risk patients [56].

Machine learning (ML) has been used extensively to solve various complex challenges in various application areas. ML can help enhance the reliability, performance, predictability, and accuracy of diagnostic systems for many diseases. This paper provides a comprehensive review of the use of ML in the medical field such as detect COVID-19. Such algorithms learn from many diagnosed samples collected from medical test reports. They can also support medical experts in predicting and diagnosing diseases in the future [90].

As an alternative, another forecast model that calculates XGBoost was developed to forecast fatality rates and differentiate the essential factors that can be determined in clinics. The researchers determined three critical factors: high-affectability C-receptive protein, lactic dehydrogenase, and lymphocyte for determining a patient's survivability. The highlight of this approach is its easy convertibility, and the triple factors recognized by the procedure are the important and critical indicators in the pathophysiological progress of COVID-19, especially cell damage, cell inflammation, immunity, and inflammation [91].

A similar study was conducted to predict whether a COVID-19 infected person may need a longer period to stay in the hospital or not, based on a U-Net AI system, which is secondarily trained using CT data [57]. While these methodologies have their shortcomings regarding scope and information, they represent important studies that can be improved with additional clinical information from other cases worldwide. Together, these methods may significantly aid in identifying infected persons needing longer stay periods at the hospital, thereby supporting hospitals in having an adequate plan.

#### *3.2. Disease Tracking and Treatment*

Computational biologists are essential in combatting the COVID-19 pandemic because of their contributions to modeling. Computational biology is called computational simulation, mathematical modeling, and data analytics for advancing biology [92]. With disease dynamics modeling, the impact of specific parameters that abet disease transmission and mediation's impact in fighting infections is better understood [93]. When a patient dies from the virus, their lungs start manifesting glass and permeating. Different data-aided medication transposition methods were developed to identify diseases, patients, or conditions that could be tackled with the medications used for different ailments [27].

#### 3.2.1. Prediction in COVID-19

As soon as a virus RNA penetrates a particular cell, it bonds the affected host cell's protein creation, using it to produce proteins replicating RNA molecules. Proteins possess a 3D structure that can be examined through sequences prearranged by amino acid order. The 3D structure affects the character and objective of the protein [61,92]. They are usually called polymerases and proteins and are the focus of treatments [94].

The two main ways of dealing with forecasts are template modeling, which forecasts structures using the same type of proteins as a framework model succession, and prototype-free modeling, which forecasts patterns for proteins with unidentified associated patterns [62]. It is proposed that these forecasts may assist in finding a cure for the COVID-19 pandemic. Further, the AlphaFold system relies on a bigger ResNet structure and utilizes amino acid order, as well as the characteristics from the corresponding amino acid order through different order structures, to predict the length and the sparsity of gradients among amino acid remains [63]. This method could be used to determine the patterns of the six proteins related to the SARS-CoV-2 layer protein, Nsp2, Nsp4, protein 3a, Nsp6, and proteolytic-like penzyme [61].

#### 3.2.2. Discovery of a Drug for COVID-19

At the Massachusetts Institute of Technology (MIT), some experts are currently developing a method for fighting the ravaging COVID-19 by producing a "decoy" receptor or protein, which might be used as a drug. The virus causes illness by attacking and attaching to the body's ACE2 receptors. The experts at MIT are using an AI concept, built on data related to ACE2 receptors, to mimic the link between the hooks and the virus [95]. Few studies are looking at ways to find new composites to focus on SAR-Cov2 by deploying novel conduits to determine constraints for the 3C-similar enzymes [64].

These systems employ three sets of data of the precise architecture of the enzyme, the c-clear substance, and the homogeneity template of the enzyme. Different types of information are used, including the productive automatic-encipher and the productive antipathetic matrix [65]. The researchers are investigating the possibility of utilizing a supplementing cognitive method with a large receptibility that can integrate factors such as the dosages of drugs, similarity, freshness, and different varieties.

#### **4. Research Challenges and Open Issues**

In this part, we have emphasized some research issues that require research consideration to attain efficient AI technologies for COVID-19 pandemic mitigation. The research issues cut across insufficient data for algorithm training, high computation expenses, security and privacy issues, and unclear interoperability functions. The detailed discussion is as follows:

#### •**Insufficient Data for Machine Training**

In some parts of the world, there are not sufficient data, such as CXR images from COVID-19-infected persons, for training the machine/algorithm. Similarly, there is no sufficient repository containing all data on the symptoms of infected COVID-19 cases. News and social media data reports may be highly unstructured, multidimensional, and low quality. Data may not be accessible from the community with limited Internet access. In addition, there are challenges in collecting patients' physiological features and therapeutic outcomes. Thus, projecting the skewed outcome of results and erroneous predictions could cause mass hysteria in the healthcare system [16,70]. Considering these challenges, there is a need to build national and global repositories for COVID-19 medical data.

#### • **High Computational Expenses**

Since different researchers have mainly employed deep learning (DL) concepts in the quest for combating COVID-19, the machine/algorithm has a high dependency on high-capacity hardware. This is because DL uses a neural network that depends on large datasets for training and testing the COVID-19 prediction model. The need for a large dataset for algorithm training also leads to a long training time, which may not be helpful for the early prediction of the virus [96,97]. Therefore, there is a need to develop more robust DL techniques that consider the urgen<sup>t</sup> need for the COVID-19 predictive model. The model should take less training time in the model training phase.

• **Scarce Data**

The primary ingredient of machine learning techniques is the large quantity of data. Labeled data are often used to train machine models to learn and make specific predictions. However, with partial data, the whole of the AI system could become flawed. Thus, the large data set might not be readily available in some countries affected by COVID-19. Therefore, there is a need for global repositories where COVID-19 patient data can be accessed.

#### • **Security and Privacy**

To restrain the transmission of the virus, mobile applications for the real-time transmission trailing, detection, and observation for quick warning and alerting have been developed. However, the privacy and security of mobile phone users is not explored [76,98]. For instance, using governmen<sup>t</sup> surveillance gadgets in public places to detect infected persons has sparked adverse reactions from people because such surveillance reveals the identity of every detected individual. Thus, there is a need to ensure that mobile phone data and other surveillance data remain anonymous in the AI technique.
