**1. Introduction**

The well-known severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2) contamination, dubbed coronavirus disease 2019 (COVID-19), has posed a universal healthcare issue. The pandemic has affected almost 215 nations across the continents, with more than 643,875,406 million confirmed cases of infection, including 6,630,082 deaths, at a rate of 1.59% deaths from all confirmed cases. However, 506,530,275 have recovered from the infection, at 78.6%, as of 9 December 2022 [1].

**Citation:** Almotairi, K.H.; Hussein, A.M.; Abualigah, L.; Abujayyab, S.K.M.; Mahmoud, E.H.; Ghanem, B.O.; Gandomi, A.H. Impact of Artificial Intelligence on COVID-19 Pandemic: A Survey of Image Processing, Tracking of Disease, Prediction of Outcomes, and Computational Medicine. *Big Data Cogn. Comput.* **2023**, *7*, 11. https:// doi.org/10.3390/bdcc7010011

Academic Editors: Domenico Talia and Fabrizio Marozzo

Received: 11 November 2022 Revised: 20 December 2022 Accepted: 23 December 2022 Published: 11 January 2023

**Copyright:** © 2023 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (https:// creativecommons.org/licenses/by/ 4.0/).

The domain of science and technology is performing an essential function in creating a cure for the virus. With the pandemic globally still raging due to the evolvement of new variants (i.e., delta variant), there has been a desperate search for ways to curtail its spread and develop a vaccine for the virus [2]. Early response to the disease in China was made by employing artificial intelligence (AI), such as tracking and tracing patients' travel history through facial recognition cameras, delivery of food and medicines using robots [3], disinfection of public buildings using drone technology [4], and dissemination of infoJessrmation to the public to remain indoors [5]. In addition, AI has been employed in the development of new molecules in the fight against COVID-19 [6], as shown in Figure 1, just as scientists are developing new drugs, along with computer experts aiming to detect people suffering from the disease via medical imaging, including CT scans and X-rays [7,8].

**Figure 1.** AI-based COVID-19 Management Architecture.

Furthermore, with the assistance of AI, tracking innovation is being developed through applications such as monitoring bracelets, which easily track patients breaching lockdown rules. The combination of AI- and mobile phone-aided cameras is also being deployed to take people's body temperature [9]. For example, the national medical insurance database in Taiwan is input with the dataset from both the custom and immigration databases to reconstruct patients' itineraries and symptoms [10,11].

Generally, AI is used to model, forecast epidemics and pandemics, diagnose [12], and validate the healthcare claims of a patient. With the help of supercomputers, different vaccines are being developed for COVID-19 [13]. In addition, drones and robots are deployed for logistics: distributing food and drugs and disinfecting public buildings.

Figure 2 represents the deadliest pandemics and data for the past 102 years. Dengue was discovered in 1950, with about a 100 million–400 million infected persons per year, which leads to about 2.5% of death. Smallpox was discovered and led to the death of about a 300 million people in the 20th century. HIV was discovered in 1920, with more than 75 million infected people and 36 million deaths. Another virus, called rabies, was discovered in 1920, with infection and death rates of 29 million and 5900 yearly, respectively. The Spanish flu was discovered in 1918, which infected more than 500 million every year and led to the deaths of about 50 million–100 million infected people. In 1973, a rotavirus virus was discovered, leading to about 0.2 million–0.5 million deaths yearly.

Afterward, an Ebola virus was discovered, which infected more than 31,000 people and led to the deaths of about 13,000 death tolls. A different virus was discovered, up until the current COVID-19, which was first discovered in 2019, with more than a 200 million infected persons and 4.4 million deaths. Several studies have been conducted to leverage the AI-centred model to enhance the COVID-19 prevention and detection process. In [14], the importance of AI was emphasized for handling the critical stage of COVID-19 prevention

and detection, which is the decision-making stage. Thus, adopting AI would double up and assist in managing patient treatment efficiently in the intensive care unit (ICU).

**Figure 2.** Deadliest Pandemics over the last 102 years (as of 25 August 2021).

Naude [10] explored several AI-related research focusing on the COVID-19 pandemic. The use areas of AI for COVID-19 include data dashboards, prognosis cures, diagnosis prediction, and tracking, warning and alert triggering [15], and social control. It is asserted that data scarcity or extensive abundant data employed in data analytics could cause an obstacle for utilizing AI for COVID-19 [16].

#### *Motivation and Literature Gap*

Several papers have proposed reviews/surveys of applications of AI for curtailing COVID-19. However, the direction and focus vary regarding the characteristics of protocols. For example, in Kumar et al. [17], improved modern technologies for handling the COVID-19 pandemic were reviewed, focusing on the functions of AI and other computer technologies for tackling the pandemic. However, related challenges and the severity of COVID-19 across different countries have not been analyzed. Further, Calandra and Favareto [18] have proposed an overview of the use of AI in combating the COVID-19 outbreak. In addition, dominant variables for AI in combating the COVID-19 outbreak were analyzed. However, current challenges concerning adopting AI for handling the pandemic have not been explored. AI application functions for fighting the spread of COVID-19 have been reviewed [19].

Similarly, a survey on AI and digital style using industry and energy for the post-COVID-19 outbreak has been proposed [20]. However, the research challenges related to security and privacy for adopting AI technologies have not been explored. Hassan et al. [21] proposed a systematic literature review for measuring the impact of AI and mathematical modeling in combating the COVID-19 outbreak. The proposal further surveyed different variants of COVID-19 and quality metrics for evaluating AI and mathematical modeling performances. However, the proposal did not look into the challenges of adopting mathematical modeling and AI paradigms.

In our proposal, we have reviewed different AI technologies and considered their impact on combating the COVID-19 outbreak. An analysis of outbreaks considering different countries is presented. Further, research challenges and open issues focusing on the application of AI for tackling the COVID-19 outbreak have also been proposed. Hence, little or no literature considered the open issues and research challenges in COVID-19 detection and control.

Considering the discussion above, this paper assesses the employment of AI in combating COVID-19. The paper comprehensively reviews the technological advancements at the forefront of the fight against the pandemic. The paper critically examines the AI-based procedures for handling COVID-19. Additionally, this paper advocates the usage of AI. In

addition, the paper explains the deployment of AI and provides context on how innovation is employed against the pandemic. Figure 3 shows the top 17 countries most affected by COVID-19.

**Figure 3.** Top 17 most affected countries by COVID-19.

The rest of the paper is arranged as follows: Section 2 involves a comparative discussion of related surveys. In Section 3, the analysis of the impact of AI technology on COVID-19 is presented. Further, Section 4 entails a discussion on open research issues and challenges. Lastly, the conclusion and recommendations are presented in Section 5.

#### **2. Comparative Discussion of Related Surveys**

This section provides a comparative discussion of the related survey, which is further divided into two subsections. Section 2.1 is about the spread of COVID-19, and Section 2.2 involves the diagnostics of COVID-19.

#### *2.1. Spread of COVID-19*

The ravaging COVID-19 pandemic has changed the direction of research studies because researchers are given more concentration on how to alleviate the virus using various techniques in the AI-centered field. In the interim, researchers have suggested reviews based on AI's function in combating COVID-19 to support relevant authorities, such as medical practitioners [12] and policymakers, in decision-making. The related surveys can be classified into problem-centered AI solutions and AI structures implemented on various COVID-19 processes.

A survey that suggested a classification of tasks involved in predicting the COVID-19 virus has been presented [22,23]. The study outlined the use area of big data and AI. However, most of the considered papers for review are not from reputable sources. In addition, open issues and current research challenges have not been highlighted in the study. In the same direction, Bansal et al. [24] precisely highlighted the function of the AI strategies employed for detecting, predicting, and controlling COVID-19 [25].

Conversely, some COVID-19 processes have not considered some parameters, such as severity assessment and death rates. Further, Kumar et al. [26] concisely extend the function of deep learning (DL) and machine learning networks to handle the pandemic, even though research studies focusing on COVID-19 treatment via respiratory waves and

clinical data have not been explored. Moreover, few studies were explored by analysis of AI-centered applications from different facets [27].

#### *2.2. Diagnostics in COVID-19*

The foundation of the AI-centered framework and big data concepts used for handling the spread of the COVID-19 pandemic have been reviewed in [28]. Discussions have been provided on the different AI-categorized learning techniques, with specific details on clinical data analysis and results about COVID-19. However, little attention has been given to analyzing the employed techniques. In a similar survey, Swapnarekha et al. [29] classified the reviewed papers into three models, i.e., ML, DL, and statistical, for handling COVID-19 and another related viruses. Further, a summarized review of COVID-19 recognition and prediction is proposed in [30].

A survey based on complicated DL has been proposed by Jamshidi et al. [31]. The survey explored DLs, such as the generative adversarial network (GAN), recurrent neural network (RNN) [32], extreme learning machine, and long short-term memory (LSTM), for a COVID-19 cure. However, the employed models are presented without critical comparative analysis. Further, a description of AI-centered forecasting and statistical model was presented in [33]. On the other hand, the only review of data mining strategies and ML for predicting COVID-19 was proposed in [34,35]. Furthermore, there was a taxonomy for complicated DL techniques for creating radiology reports [36].

Several studies reviewed a certain kind of dataset; for instance, Jalaber et al. [37] put forth the function of CT images for handling COVID-19-infected patients.

The function of the CT scan was also used for handling the presentation of lesions and severity signs. At the end of the paper, five related papers were explored to describe the AI's function for COVID-19 diagnosis. The landscape of radiographic imaging structures and AI methods was investigated. The imaging structures, such as PET, CXR, and CT, were considered for the AI data training and testing. However, the papers considered had constrained information regarding the gained results [38]. In another survey, the imaging characteristics of PET-CT and CT from several articles were presented [39], as well as a comparison of the AI techniques applied for COVID-19 prediction [40]. AI methods for diagnosing COVID-19 have been discussed by categorizing CT and CXR images [41]. In both [42,43], the domain of biosensors and IoT for handling the COVID-19 pandemic have been discussed.

Our paper surveyed articles containing AI's concept for handling COVID-19, in terms of prediction, diagnosis, survival assessment, drug discovery, recasting, and pandemic outbreak. Considering the discussion mentioned above, it is evident that a study focused on a distinct part of COVID-19 handling or described a single type of dataset. Further, many of these reviews offered fewer relative analyses and examined few papers. Conversely, there are a handful of articles that have not been surveyed.

The following section explores and presents the impact of AI in handling COVID-19.

#### **3. Impact of AI on Repressing COVID-19**

This part discusses the use of artificial intelligence (AI) techniques for handling the COVID-19 pandemic that have been discussed. AI technologies could be based on natural language processing (NLP), ML, and other applications of computer visualization. The different capabilities allow machines to use large information-based frameworks to build, show, and foretell. Table 1 presents numerous uses of technology in the fight against COVID-19. AI is often used to diagnose viruses, analyze medical images, trace, track, and carry out future disease predictions [44]. In addition, it is also used to send alerts to raise awareness and create social awareness virtually.


**Table 1.** Use cases of AI in CT diagnosis for COVID-19 pandemic.

AI techniques have been used to extract the exact graphical features of COVID-19 that help provide clinical treatment/diagnosis before conducting the pathogenic test, thereby minimizing the time for pandemic control. By employing radiology images in diagnoses, AI obtains radiological characteristics for the prompt and precise discovery of COVID-19 [45]. The techniques employ deep learning algorithms using a computer vision model that considers specific parameters, such as level of specificity, accuracy, sensitivity, region area under the curve (AUC), negative predictive value, and positive predictive value. Similarly, deep convolutional neural networks (CNN), which employ X-ray image data for model training and testing, have been proposed for the automatic detection and prediction of COVID-19 [54]. The proposed technique serves as a substitute treatment and diagnosis decision to avoid the spread of the coronavirus among the infected people around the globe using CNN-based models, which include ResNet50, ResNet101, ResNet152, Inception-ResNetV2, and InceptionV3 [54].

Another solution Wang et al. [49] proposed for handling the pandemic is COVID-Net. It employs the AI concept for detecting coronavirus using the data from an open-source repository of chest X-ray images. Another AI technique has been proposed to screen coronavirus using multiple CNN to classify images and find the probability of the virus infection [47]. The current CT application, and/or the above AI techniques that have been proposed, appear to help ascertain the pandemic to provide diagnosis/clinical support to a patient before conducting the pathogenic result that is ready for proper action.

Wang et al. [55] presented a somewhat effective respiratory simulation model (RSM), in order to handle the limitation between the massive volume of training data and the limited available real data. Meanwhile, the suggested deep learning model could be expanded to big-scale use areas, such as office environments, sleep scenarios, and public places. Although, the technique has faced some challenges, including adequate real-world data to realize the learning method, and the variation in different respiratory patterns is also less than average. The disease tracking procedure involves the following steps: (1) irregular respiratory sequence classifier, which can lead to mass testing of people infected with COVID-19. (2) The SIR model, which is time-bound, is employed for determining the number of infected people. (3) The gated recurrent unit (GRU) is a neural network that uses an embedded bi-directional and attentional system (BI-AT-GRU) for categorizing respiratory sequences. (4) The infectious, exposed, vulnerable, and eliminated or recovered framework is employed to predict the cause of the pandemic.

An ML-based model for predicting the survivability of patients infected with COVID-19 and the prediction result of the patient's state of health has been presented in [56,57]. In [56], the supervised XGBoost classifier gives a straightforward and spontaneous medical screening to measure the likelihood of bereavement accurately and promptly. In [57], the ML-based CT frameworks indicated the possibility and precision of forecasting the stay time of patients infected with COVID-19 at the hospital.

A model is a supporting tool for decision-making and logistical planning for the healthcare system. The technique uses different algorithms with different datasets. Richardson et al. [27] also employed Benevolent AI's knowledge graph to search for approved drugs to help minimize coronavirus infection. The authors did not discuss the detail of the algorithms and how the model performance was evaluated using the available parameters. Similarly, a novel deep-learning pipeline architecture has been proposed [58] as an alternative to COVID-19 detection. The technique uses a chest x-ray image with convolutional CNN to detect whether the patient is a carrier of COVID-19 or not, with detailed diagnosis features and a quicker diagnosis. The technique has been regarded as the most suitable in places that have advanced computing machines. However, during this pandemic, people need a solution that can be integrated with existing and/or available resources. Another technique, based on a cuckoo search optimization algorithm, has been proposed to extract basic information from the X-rays conducted on the lungs using three classification processes: called normal patients, COVID-19-infected patients, and pneumonia patients.

The approach is an alternate solution to detect COVID-19 from the X-ray images using a modified CS algorithm [59,60].

Protein structure prediction is used to extract some features from medical images. In [61], the residual learning procedure was utilized to simplify the training of considerably deep systems for image feature detection. In [62], the critical assessment of methods for protein structure prediction (CASP) by employing a deep neural network to forecast protein characteristics based on its genetic pattern was suggested. In [63], convolutional network architecture was inspected for heavy projection.

Drug innovation is an application for adversarial auto-encoders, which is employed in extracting the method and the structure of image data, dimensionality reduction, unsupervised clustering, and data conception [64]. While in [65], protein structure is used as an incorporated AI-centered drug detection conduit to award new drug mixtures.

In [66], cough-type diagnosis utilized a considerable selection of acoustic characteristics administered to the documented audio from many uninfected and infected persons. In [30], a smartphone thermometer was a simple substitute device for measuring the temperature of infected persons.

Social media has become very popular worldwide, as it is used to interact and communicate [25]. However, one problem is information overload, misinformation, and fake news. To counter this "infodemic", the World Health Organization (WHO) introduced the information network for epidemics (EPI-WIN) to distribute news and data with some major partners [67]. Social media giant Facebook analyzes posts about infections; its ad library [68] examines all ads through the tag "COVID-19" and "coronavirus", and Facebook aggregated 923 outcomes in 34 nations, the maximum of which were from the US (39%) and Europe (Italy had 25% of the ads).

A system for detecting COVID-19 utilizing data from mobile phones' sensors, including cameras, microphones, inertial sensors, and temperature, was proposed in [66]. Similarly, audio data collected from hand-held phones' microphones was employed to identify coughing [30]. It is essential for AI to be trained to predict infection threats and, as such, help identify high-risk cases for containment purposes, thereby curtailing the spread of the virus among the populace [69]. Some drones were also used to trail and detect people who were not using mouth/nose masks, and some were employed as a public address system to address the public or disinfect public places. A company from Shenzhen in China, Small-Multi-Copter, has helped dramatically with logistical support and distribution of medical supplies and lockdown materials via drones.

To curtail the transmission of the virus in India, the authorities introduced Aarogya Setu [70], a mobile phone app that could track coronavirus patients to fight the infection on an individual basis. The app also helped trace contamination using mobile phone GPSs and Bluetooth to collect data on whether a person has come into contact with a COVID-19 patient. To curtail further infection from the coronavirus in India, the authorities developed a mobile application known as Aarogya Setu [70], which tracks coronavirus infection and also aids in stopping the spread from person-to-person. It aids in tracing coronavirus infection by using mobile phones' GPS networks, as well as the Bluetooth of the phones, with which it detects whether an individual has had an interaction with a COVID-19 patient [71].

#### *3.1. Medical Image Processing*

The effectiveness of the current diagnostics at the beginning of the pandemic was challenged. Open clinical methods were ineffective against the COVID-19 virus, and with limited medical equipment and other assets, the cure needs of every patient were determined by the seriousness of their symptoms. With many outpatients with mild symptoms that could suddenly be serious, there was a need to diagnose the symptoms early enough for effective treatment and ultimately drive down the mortality rate. Therefore, AI could be effective in the prognosis, prediction [72], and curing of COVID-19 patients and drive down treatment costs [73]. Most medical uses of AI are often used for diagnosis using

medical imaging. In some current studies, it was established that only a small number of the studies used AI in arriving at their CT scans. Additionally, other studies employ patients' medical records to predict the severity of the virus [5,45,54].

#### 3.1.1. The Role of CT Scan for COVID-19 Patients Screening

The results obtained based on CT from COVID-19 scenarios include multilobar GGO and bilateral with surface or subsequent distribution. This is often in the lower lobes and with less frequency in the intermediate lobe. Subpleural, septal thickening, pleural thickening, and bronchiectasis involvement are a few of the usual outcomes, particularly in the subsequent phase of the virus. CT halo symbol, pleural effusion, lymphadenopathy, pericardial effusion, cavitation, and pneumothorax are also among the few unusual, but probable, outcomes observed from the virus evolution [3,6,7].

Bai and his team noted common features within 201 infected patients, CT irregularities and suitable RT-PCR patients, as follows: 80%, 91%, 56%, and 59% for the surface circulation, GGO, good reticular opacity, and vascular congealing, respectively. Fewer usual features for the CT photographing on the chest included the following: 14%, 2.7%, and 4.1% for the central and peripheral distribution, lymphadenopathy, and pleural effusion, respectively [10]. At an early stage, the chest film is not usually sensitive, and it can be found to be significant at a later stage during the monitoring of the disease [11]. According to Malpani et al. [74], another way of calculating the severity score is to assign the percentages of individuals of the five given lobes shown as <5% contribution, 5–25% contribution, 26–49% contribution, 50–75% contribution, and >75% contribution [8,10]. The overall CT mark contains the summation of each of the given lobar marks that cover the bound of values from 0 to 25 (for no contribution and maximum contribution, respectively), once the contribution of all the five lobes is found to be above 75% [11].

#### 3.1.2. Diagnosis Using Radiology Images

With the application of AI, many lives could ultimately be saved, and the spread of the virus could be checked, leading to the generation of relevant data from AI models with correct diagnoses of the virus. With AI, radiologists could achieve faster, not to mention cheaper, diagnosis rates than mainstream coronavirus tests [75]. In the same vein, doctors could also use a combination of X-rays and CT scans [46]. The different AI use cases for handling COVID-19 are presented in Table 1. COVID-19 medical tests are not widely available and often expensive, but most emergency and trauma clinics usually have CT and X-ray machines. Thus, with the help of DL, a radiology expert could analyze and detect the presence of COVID-19. In another development, COVID-Net has proposed an IT-based application for examining COVID-19 signs, based on CXR, through the various information of the lungs of infected patients [76]. Using diagnostic research, AI software was developed from an inceptions migration neural network for analyzing and detecting COVID-19 symptoms with the help of CT images with an 89.5% accuracy rate [45].

A preliminary discovery model has been developed to detect the COVID-19 virus from Influenza-A and specific cases with pulmonary CT images using a DL system. The affected portions of the patients were aggregated using the 3D DL model, and the research had an 86.7% accuracy rate [47,48]. Similarly, Cao et al. [77] built a DL system to effectively diagnose the virus symptoms contracted from other lung ailments and community-acquired pneumonia (CAP). Using chest CT scans, a 3D learning method was developed using a DNN (COV-Net) [78]. Additionally, to diagnose coronavirus, a DL framework was developed that quickly uses CT data as inputs, carries out lung categorization, detects COVID-19, and diagnoses any irregular slice. In addition, the research shows that the diagnoses of AI methods could be explained using data to check the shortcoming of the DNN model as a black box [79]. Subsequently, a computerized system has been developed to quantify the different signs of the virus in patients' lungs and check the virus or response to treatment by employing a DL technique. The range of capabilities of AI clinical analyses have not ye<sup>t</sup> been determined. However, some hospitals in China

have been using AI-aided radiology innovations. Transcription polymerase chain reaction (RT-PCR) tests are critical for diagnosing coronavirus. However, they have their limitations regarding specimen variety and the duration needed for the research and processing [50]. Some abnormalities in CT image data of COVID-19 have been seen using the central AI concept [80,81]. Similarly, the fuzzy-based decision-making technique has been explored by [51] to assess the severity of COVID-19 in the Kingdom of Saudi Arabia (KSA), while adopting a more robust computational model for evaluating the severity using social influence to control the spread of the COVID-19 pandemic [51].

An X-ray technique with an automated system identifier for COVID-19 detection on chest images has been designed using the convoluted CNN architecture. The technique extracts the feature descriptors from the chest X-ray image using a speed-up feature robust algorithm and integrated k-means clustering algorithm to detect whether there is a presence or absence of COVID-19. The study used the dataset of 340 X-ray radiographs and 170 images of both healthy and positive COVID-19 classes [52]. Chest X-ray (CXR) has been used to detect the coronavirus infection by proposing the dilated CNN, branching design model, and VGG-16 technique. Therefore, the VGG-16 used the beginning of the ten layers in the model's front end to extract and utilize the high-level merits [53].
