A Survey on the Use of AI and ML for Fighting the COVID-19 Pandemic

I Introduction

The novel and contagious viral pneumonia, COVID-19 (Coronavirus disease-2019) has affected more than 8.2 million people and caused death of more than 440,000 thousand people worldwide. WHO declared it as a global pandemic[36] and suggested that early detection, isolation and prompt treatment can be useful to slow down the COVID-19 outbreak [35]. Therefore, various bodies have committed themselves to conduct research focusing on COVID-19 to support global response.

With new discoveries being announced at a breathtaking pace, artificial intelligence (AI) has re-emerged into scientific consciousness. AI is a branch of computer science that can be used to build intelligent systems. It is often instantiated as a software program [42]. Recent application of AI in diagnosing disease s has broadened the frontier of AI which once was a humans’ expertise. Medicine and the health care systems are among the most promising areas of application of AI which can be traced back to as early as mid-twentieth century[1]. Researchers proposed and successfully developed several decision support systems[30]

. The rule-based system gained success in the late ‘70s

[49] and had been useful in detecting disease[9], interpreting ECG images[22], choosing appropriate treatment and generate hypothesis by the physicians[29]. Unlike this first-generation knowledge-based AI system, which relies upon the prior medical knowledge of experts and the formulation based rules, modern AI leverages machine learning algorithms to find pattern and associations in data [10] [45] [46]

. The recent renaissance in AI can be attributed to the successful application of deep learning to a great extent- training an artificial neural network with a large number of labelled datasets. A modern deep learning network usually contains hundreds of hidden layers

[56]. The recent resurgence of AI has been fueling a question of whether AI-doctors will replace human physicians shortly. Yet to be confirmed, researchers believe that AI driven intelligent systems can significantly help human physicians in making better and quick decisions and even sometimes remove the necessity of human decision i.e. radiology[17]. The increasing data in health care resulting from increased use of digital technology and the advancement of big data analytics can be attributed to the recent success of AI in healthcare[17]. Although AI research in healthcare is emerging, most of the research is concentrated on three diseases: cancer, neurology and cardiology. Guided by evidence, a strong AI can reveal the insight of the medical data which eventually can be used for decision support and forecasting[31], [21], [11].

As AI has been proved useful in healthcare, researchers suggest that it may also be helpful in fighting against COVID-19. From forecasting of a pandemic to designing anti-viral-replication molecules, AI has made a paradigm shift in health care. Recent research on COVID-19 using AI suggests that AI can be helpful in detecting COVID-19 infection, detecting infected population, predicting the next outbreak, finding the attack pattern and even finding a cure [16], [52] , [41]. The objective of this review is to explore the existing AI-based research that has been conducted to fight against COVID-19 pandemic. The organization of the remaining section is as follows. The methodology to conduct this review study is discussed in section 2. The review data analysis and findings are discussed in detail in section 3. Section 4 presents the main findings and the potential scopes of future research to fight against the COVID-19. Finally, the concluding remark, research limitation and future work are presented in section 5.

Ii Methodology

In this research, a systematic literature review procedure [48] was adopted to attain the research aim. For selecting the primary articles, the major databases such as IEEE Xplore, Springer Link, ACM digital library, Science Direct, and Google Scholar were searched for related articles. The search strings used to find the literature were “Machine learning and COVID-19”, “Machine learning and Coronavirus”, “Artificial Intelligence and COVID-19”, “Artificial Intelligence and Coronavirus”, “Artificial Intelligence and lockdown and pandemic and COVID-19”, “Machine learning and lockdown and pandemic and COVID-19”, and “Coronavirus prediction and outbreak prediction and machine learning and artificial intelligence”. These strings were applied for all the above-mentioned databases as well as Google search engine.

The search returned more than 634 articles. The exclusion criteria include (a) duplicate articles that are found through several scholarly databases; (b) articles that are not focused on our research objectives; (c) the article was written other than English; and (d) the earlier version of any article that has been published on the same set of data and explored the same objective. After applying this inclusion-exclusion process, we finally selected 35 articles that include original research, review articles, and short articles including perspective, commentary, and letter to the editor. The Prisma flowchart in Figure 1 shows the article selection process in different phases following the exclusion-inclusion criteria.

The selected articles were reviewed systematically to extract data primarily related to the article type, publication time, research objectives, study context, study outcomes/findings, methodology/algorithm/techniques used, dataset used, and study subject. Finally, the extracted data were synthesized and analyzed to summarize the existing research on COVID-19 pandemic using AI and to identify the potential scopes for future research.

Fig. 1: Article inclusion and exclusion process flowchart

Iii Data Analysis and Findings

Iii-a Type of Publications

To date (12th June, 2020) out of 35 articles, 27 (77%) were published as original research. Among others (23% articles), three articles were the review articles, two were editorial and the remaining three were published as research perspective (short conceptual article). Again, 21 (60.0%) articles were published in academic journals, while 14 (40%) articles were archived as pre-print. Among the pre-print, 92.85% (13) were original research. All the selected articles were published or archived in the online databases during January-May 2020.

Iii-B Research Purposes and Objectives

We synthesise the existing research in terms of their purposes and aims to explore the contribution of AI to fight against the COVID-19 pandemic. A summary of the synthesized data is presented in Table I

to show the research scopes and purposes of the original research. Most of the articles (n =16, 48%) were published focusing to detect the COVID-19 infected patients using different AI-based algorithms that include, for example, the Convoluted Neural Network (CNN) model, Support Vector Machine (SVM), generative adversarial network (GAN), and the transfer learning. The chest X-ray images, CT images, mobile sensors data, and COVID-19 symptoms were used to predict/detect the COVID-19 patients. These researches aimed to identify, screen and detect the COVID-19 patients; and also to predict, differentiate, or classify the patients into COVID-19 infection, no infection, and other viral or bacterial infection. For example, Wang et al.

[50]

proposed a CNN based prediction system named COVID-Net that identifies non-COVID-19 infected, COVID-19 infected, and no infected patients using chest X-ray images. The proposed model was pre-trained on ImageNet (open source) dataset and then trained on the COVIDx (author- created) dataset that includes 13,800 chest X-ray images of 13,725 patients that includes 183 images from 121 COVID-19 positive patients, 8066 images are of healthy patients and 5538 images are of non-COVID19 patients.

Purposes	Brief Description	Reference	Frequency
	Identify the infected individual quicker	[41]
	Screen coronavirus diseases using deep learning	[8]
	Identify the coronavirus patients	[13]
	Developing a CNN based Algorithm to detect COVID-19 from CT images	[52]
	Detect COVID-19 with the help of AI and smartphone sensors	[26]
	Use an anomaly model based on deep learning network to make the screening process faster for Covid-19 detection from X-ray images	[50]
	Detect COVID-19 from X-ray images using transfer learning with CNN	[4]
	Detect COVID-19 from X-ray images using deep CNN model	[33]
	Propose an algorithm to detect COVID-19 from CT images using deep CNN model and SVM classifier	[37]
	Develop a deep learning model CoroNet using the Xception CNN to detect Covid-19 from Xray images	[20]
	Build a framework which uses smartphone sensors to detect Covid-19	[26]
Diseases detection	Classify patients in to non-COVID 19 infection, COVID-19 infection, and no infection from X-ray images using deep CNN model	[50]	16
	Compare the performance of seven DL model to find the best model for Covid-19 detection	[5]
	Develop and evaluate the performance of an AI model to detect Covid-19 and also evaluate the performance of radiologist to detect the disease by using and without AI support	[6]
	Detect the Covid-19 by identifying the characteristics from chest X-ray using a deep learning model(CAD4COVID-XRay)	[32]
	Detect COVID-19 from X-ray images using Generative Adversarial Network (GAN) and Deep learning transfer	[25]
	Diagnosis the identified patients to classify (in to patients’ categories) and tracking the progress Covid-19 patients	[14]
	Distinguish COVID-19 from pneumonia using Deep learning	[24]
	Efficiently diagnosis COVID-19 using X-ray images through deep CNN models	[15]
	Developing a tool to predict survival and death for severe COVID-19 patients	[54]
Diseases diagnosis	Diagnosis Covid-19 positive case faster using both non-image and image clinical data	[28]	7
	Develop a system to identify patients who would develop more severe illness among the patients with mild cases of COVID-19	[18]
	Develop a system to improve the diagnostic performance from Posterior-Anterior (PA) X-ray images of lungs with COVID-19 cases	[12]
	Forecast of the COVID-19 to estimate size, lengths and ending time of COVID-19 across China	[16]
Epidemic forecasting	Predict the trend of the infection for the next 80 days using deep learning	[55]	2
	To predict the progress of the epidemic (epidemic sizes and peaks)	[55]
Sustainable development	Analyze the correlation among environmental factors and confirmed cases of COVID-19	[38]	1
	Compare the prediction performance of the proposed algorithm with the existing methods	[37]
	Compare seven different DL model to find out the best model for disease detection	[5]
Performance comparison	Compare the performance of radiologist in distinguishing Covid-19 from other pneumonia with and without AI assistance	[6]	4
	Compare the performance of a DL model with six other radiologists	[32]
Patient management	To improve management of Covid-19 ICU patients	[39]	1

TABLE I: A summary of the literature based on their purposes

A total of seven (20%) articles focused on diagnosing COVID-19 patients through AI (Figure 2). In these articles, the AI was used to diagnose the identified COVID-19 patients to classify in to patients’ categories (severe, mild) and tracking their progress[13], distinguish COVID-19 from pneumonia [24], efficiently diagnose COVID-19 using X-ray images[15], predict survival and death for severe COVID-19 patients[54], identify patients who would develop more severe illness [18], and to estimate uncertainty in Deep Learning solutions to improve the diagnostic performance from Posterior-Anterior (PA) X-ray images of lungs with COVID-19 cases [12].

Fig. 2: Percentage of papers based on objectives

Two (6%) articles aimed to forecast the COVID-19 epidemic to estimate the progress of the epidemic in terms of its size, lengths, peaks, and ending time as well as predict the development trend of the epidemic for the next certain time period in a specific country or geographical region [16][55]. We found only one study [38]

focused the sustainable development that analyzes confirmed cases of COVID-19 through a binary classification using AI and regression analysis and explores the correlation among confirmed cases of COVID-19 in four countries (China, Italy, South Korea, Japan) and environmental factors (low, high & average temperature, humidity, wind flow). An article

[37] compared the prediction performance of the proposed algorithm with the existing VGG-16, GoogleNet, and ResNet50 method using two different subsets of data, while in [39], an AI-based model is developed based on the existing studies regarding AI in ICU and respiratory diseases to improve the (COVID-19) ICU patient management.

The remaining (26%) articles include review articles, editorial, perception, commentary, and short communication. The review articles are summaries of the existing research with the aim to highlight the contributions and constraints of AI [34], and to identify a roadmap of AI applications to fight against COVID-19 pandemic [7]. Another review analyzes the AI-based techniques used in the CT and X-ray based medical imaging to fight against COVID-19 pandemic [48]. One of the two editorials highlighted how AI-based solutions may assist to fight against the pandemic by forecasting the pandemic to design anti-viral replication molecules, but with the supervision of humans [27]. In another editorial, a workflow is presented to highlight the processes and applications of AI to fight the COVID-19 pandemic [2]. The perspective articles highlight firstly the needs of AI and the ways of data sharing (via smart city networks) for better monitoring and management of urban health on the COVID-19 outbreak [3]

; secondly, the importance of active learning-based AI tools for coronavirus outbreak

[44]; and finally, suggested how AI and Blockchain can be used to help the community during the COVID-19 pandemic with equipment and donations [19]. By using a private blockchain network to make donations for the pandemic, there would be no alterations and the donations would go to their destinations. A summary of the synthesized data is briefly presented in Table II to present the research scopes and purposes of the other types of research.

Purposes	Brief Description	Reference	Frequency
	Review the related work to highlight the contributions and constraints of AI in fighting the Covid-19 pandemic	[34]
Review Literature	Review related work to identify a roadmap of AI applications to fight against the pandemic	[7]	3
	Review the AI based techniques used in the CT and X-ray based medical imaging data acquisition, segmentation and diagnosis to fight against COVID-19 pandemic	[48]
Editorial	Highlight how AI based solutions may assist to fight against the COVID-19 pandemic	[27]	2
	The editorial constitutes existing works, current efforts and potential work ideas to fight against Covid-19 using AI, ML algorithms, deep learning, neural networks.	[2]
	Highlighted the needs of AI and the ways of data sharing via smart city networks for a better monitoring and management of urban health on the COVID-19 outbreak	[3]
Perspective	Discussed the importance of active learning based AI tools for coronavirus outbreak. Tools that use cross population training/testing methods and multitudinal and multimodal data.	[44]	3
	Introduced AI and Blockchain and suggested how they can be used to effectively help the community with equipment and donations.Introduced AI and Blockchain and suggested how they can be used to effectively help the community with equipment and donations.	[19]

TABLE II: Scopes of other types of research

Iii-C Context of Study

Some of the articles focused their research on specific countries while others conducted research with a global perspective. A total of 10 articles (29%) focused on specific country as shown in Table III. One of these articles considered confirmed cases from 34 provinces of China for their research on a forecasting system of those areas [16]. Another article focused on 42 provinces in Japan, China, South Korea and Italy for environmental parameters, weather trends and confirmed cases to measure correlations and also build a classification model [38]. In one study CT scan of lungs from patients of both USA and China [13] were used. In another study, CT scan of lungs only from China [52] were used for training and testing automated AI-based tools for diagnosis and tracking. Epidemiological data of three provinces of China (Hubei, Guangdong and Zhejiang), SARS 2003 epidemic data of all over China were collected and the prediction was made for the whole China [55]. CT images of patients of Italy were used as well in another study [37]. Data was collected from only Wuhan, China in one of the articles [54] and Wenzhou city of Zhejiang province in another article [18]. As we see most articles concentrated on data from China as it is the original epicentre of the pandemic.

Fig. 3: A brief on the context of data used by the literature

Contextual articles focused mainly on epidemic forecasting and sustainable development. Most of the disease detection related articles and all of the recommendation type articles used global perspectives as well as public datasets and are not context-sensitive (Figure 3). Thus, it can be said that the disease detection techniques mentioned in the articles are mostly not context-dependent; and for the epidemic forecasting purpose researchers need contextual data. There were two cross-country studies - one of them [38] focused on finding correlation among multiple countries’ COVID-19 cases and the other study [13] focused on two different country cases to enhance the performance of their disease detection tool. Table III briefly shows the details on the data used in these contextual studies.

Literature	Objective	Data Source	Data Volume	Data Type
Hu et al. [16]	Epidemic forecasting	WHO and local Chinese news media collected Data	15,384 and 36,602 cases Clinically confirmed and lab confirmed cases respectively	Time series data (Non -Image)
Pirouz et al. [38]	Sustainable development	Data from 42 province of China, Japan, Italy and South Korea	-	Environmental, geographical and demographical data from 28 January 2020 to 26 February 2020(Non -Image)
Gozes et al.[13]	Diseases diagnosis	Development Dataset Source: Chainz Testing Dataset Source: Hospital in Wenzhou, China, Chainz, El-Camino Hospital (CA), LIDC Lung segmentation Development Sources: El-Camino Hospital (CA) sources were used	157 patients	CT scan images of lungs(Image)
Wang et al [51]	Disease detection	China	453 images from 99 patients	CT images of chest (Image)
Yang et al. [55]	Epidemic forecasting	Covid-19 outbreak data reported by the National Health Commission of China(Wuhan, Hubei province, Guangdong province, Zhejiang province) , Migration data was retrieved from a web based program, 2003 SARS epidemic data was retrieved from an archived news-site (SOHU)	-	Non -Image
Umut et al. [37]	Disease Detection	Societa Italiana di Radiologia Medica e Interventistica (Itali)	150 CT images	Time series data(Non -Image)
Li et al.[54]	Diseases diagnosis	Wuhan (China) clinical Data	3129 cases of COVID-19 patients	Time series(Non -Image)
Xiangao et al. [18]	Disease Detection	Clinical data from Wenzhou, Zhejiang, China.	53 hospitalized patients’	Medical data (Non -Image)
Mei et al.[28]	Disease Diagnosis	Chest CT studies and clinical data from China	905 patients	Chest CT images And clinical data (Non-image)
Harison et al. [6]	Disease Detection	Chest Xray from Hunan province, China	512 patients	Chest X-ray (Image)

TABLE III: A brief details on Data of the contextual literatures

Iii-D Exploration of the Used Data Type

24 studies (68.5%) had used different types of data, ranging from text to image, to corroborate their findings as shown in Figure 4. The higher percentage (63%) with 22 original studies prove the dependency of AI and ML based systems on proper data assessment.

Fig. 4: Analyzing the Data used in the reviewed articles

Seventeen (40%) of the articles used image data in the form of X-Ray images and CT images of the chest. These images are mostly used in disease detection and diagnosis. Seven studies (23.3%) used public datasets among which four studies used the COVID-19 dataset from GitHub repository created by Dr. Joseph Cohen, a postdoctoral fellow at the University of Montreal. X-Ray images for other lung disease patients, such as pneumonia, were collected from GitHub repository, Kaggle repository and Open-I repository [43]. Two of the other studies collected data from various hospitals from China and the USA (Table IV). One study collected data from Societa Italiana di Radiologia Medica e Interventistica [26], a hospital from Italy.

The other seven articles (20%) used non-image data, predominantly in the form of text and numbers with the purpose of disease detection, epidemic forecasting, sustainable development, introducing advanced concepts, and disease diagnosis and progression (Table I). Thousands of data points were collected through a mobile survey for a study [41] that included information related to location, age, gender, race, travel and close contact to any affected person. Furthermore, the study collected health data related to COVID-19 symptoms during a period of 14 days. Clinical data including information on baseline characteristics, medical history, COVID-19 diagnosis from hospitals in Wuhan and other provinces in China were collected for disease prediction and progression purposes. The study [38] used environmental and urban data accumulated from 42 different provinces in China, Japan, South Korea, Italy to analyze the correlation among environmental factors (low, high and average temperature, humidity, wind flow) and confirmed cases of COVID-19. The study here [55] used migration data and 2003 SARS epidemic for epidemic forecasting of COVID-19. Finally, the study[2] used various research papers and scholarly articles for the purpose of proposing potential ideas to combat COVID-19 using AI.

We observed that nine (25.8%) studies used data that were collected from different provinces in China making China as the major source of initial COVID-19 related Data.

Iii-E Exploring the AI Techniques

Most of the research papers (n=29, 83%) aim to use AI to do some kind of classification (COVID-19 detection, differentiate COVID-19 from other respiratory diseases) forecasting, and prediction (Table I). The cost and time associated with the gold standard of testing COVID-19: PCR - takes up to two to three days to get the results, drive researchers to find an easier, cheaper and faster way to detect COVID-19 using computational technique. Therefore, our study found most of the research work (65%) aimed to detect and diagnose COVID-19. Table IV briefly presents the objective, scope and the results of using different AI algorithms.

Fig. 5: The percentage of literature used different data type for different algorithm

COVID-19 has put researchers, health professionals at a critical situation due to the lack of timely information and historical data. Intelligent systems cannot work unless they are trained with reliable data. Application of AI and other related techniques: machine learning, deep learning is done based on the previous experiences i.e data and models. Given that very little information related to COVID-19 are available, researchers mostly rely on X-ray images and CT scan images. Although a very small amount of Chest-X Ray (CXR) of COVID-19 is available, CXRs are prescribed as one of the first diagnostic tests by the physician. Most of the earlier research works use CXR images to detect COVID-19 (26.67%) and others use chest CT images (Figure 4). The recent development of deep neural networks has opened up a new frontier in image classification. We found most of the research papers (43%) use different architecture of deep neural networks (Figure 6) to classify images, both CXR and CT scan (Table V). When it comes to image data, Convoluted Neural Networks dominated over all other algorithms and techniques(Table IV). Using CNN as a base, several studies come up with their architecture [50]. Our observation finds out among several CNNs, the Res-Net architecture as the most used one (Figure 5

). Using the Res-Net architecture as a backbone, the models are different from one another on several parameters i.e. several hidden parameters, epochs and optimizer. Some studies also use a combination of different deep neural architectures. For example, Ezz et al.

[15] compared six other different models with Red-NetV2 to propose the best one. Some studies, for example, Wang et al. [51]

use a combination of Res-Net with other machine learning algorithms. The authors used Res-Net for feature extraction from CXR and leverage SVM and decision trees to develop a new algorithm. Other than classification, Biraja et al.

[12] use different architectures i.e. bayesian CNN, monte Carlo drop weights along with Res-Net to assess the uncertainty associated with applying deep neural networks to detect COVID-19. Due to the pervasive use of deep learning networks and models, we have used a separate a table, Table V to present the deep learning architectures used and their results.

Fig. 6: Literature frequency of different AI techniques

Other than deep learning approaches, traditional machine learning algorithms have also been applied when it comes to non-image data (Table IV). Using a combination of regression analysis and Group Method of Data Handling(GMDH), Pirouz et al. [38] tried to find a correlation and forecast based on demographic factors. Moreover, Yang et al.[55] combined epidemiological models with an ML model to show the effectiveness of the disease containment in China and predict the epidemic. With an addition to this, Loey et al.[25] used neural networks (GAN network) for detecting COVID-19, whereas Jiang et al.[18] used decision tree, SVM based algorithm to detect COVID-19 from Xray images.

The authors depend on the train-test split method for validation of their models as none of these models is used to test on real patients. As a validation metric, the studies have used accuracy, specificity, sensitivity, f-1 score and area under Receiver Operating Characteristic(ROC) curve (AUC). Among other evaluation matrics authors also used False Positive Rate (FPR), True Positive Rate (TPR), positive and negative predictive values (PPV and NPV respectively). Our finding suggests that deep learning algorithms achieve a higher score in most of the evaluation matrices (see Table IV, and V).

Objectives	Algorithms	Evaluation Results	Literature
	AI- based algorithm	-	Rao et al.[40]
	CNN	Accuracy (82.9% ), Specificity (80.5%), Sensitivity (84%)	Wang et al. [52]
	CNN	Accuracy (97.8%)	Apostolopoulos et al. [4]
	CNN	F1-score (0.89)	Ezz et al. [15]
	CNN	Sensitivity (100%), Specificity (100%), Accuracy (100%), F1- score (100%)	Salman et al. [43]
Disease detection	CNN	Accuracy (98%), Recall (96%), Specificity (100%)	Ali et al. [33]
	CNN, SVM	Accuracy (98.27%), Sensitivity (98.93%), Specificity (97.60%) F-1 score (98.28%), Precision (97.63%), Matthews Correlation Coefficient (96.54%)	Umut et al.[37]
	CNN,SVM	Accuracy(95.38%), FPR(95.52%), F1- score(91.41%) , Kappa (90.76%)	Sethy et al.[47]
	GAN Network	Accuracy (99.9%)	Loey et al. [25]
	Decision trees, random forests and support vector machines	Accuracy(80%)	Xiangao et al. [18]
Epidemic forecasting	Modified Auto-encoder for Modeling Time Series	-	Hu et al. [16]
	Epidemiological model and ML based AI model	-	Yang et al. [55]
Sustainable development	Regression analysis and Group method of Data Handling	Accuracy (85.7%)	Pirouz et al. [38]
	CNN and Grad Cam	AUC (0.989), Sensitivity (98.2% )	Gozes et al.[14]
Diseases diagnosis	CNN	AUC (0.96)	Li et al.[24]
	XGBoost machine learning algorithm	Death prediction accuracy (100%), Survival prediction accuracy (90%)	Li et al.[54]
	CNN	-	Biraja et al. [12]

TABLE IV: A summary of the algorithm used in the literature for different objectives

Literature	Architecture	Task	Result	Research Outcome
Gozes et al. [13]	U-net, Resnet-50-2D	Classification ,quantification and tracking: Covid-19 patients	0.996 (AUC) 98.2% (Sensitivity) 92.2% (specificity)	AI based software
Wang et al [51]	ResNet-18	Feature extraction from image data	73.1% (Accuracy) 67% (specificity )and 74% (sensitivity)	A CNN based algorithm leveraging decision tree and SVM
Li et al. [24]	ResNet-50	Classification : Covid-19 and pneumonia	0.96(AUC)	A CNN based model: COVnet
Sethy et al. [47]	Resnet50	Classification : COVID-19	95.38%(Accuracy), 95.52%(FPR), 91.41%(F1- score) , 90.76%(Kappa)	A CNN based model
Apostolopoulos et al. [4]	VGG19 , Mobile Net ,Inception ,Xception Inception ResNet v2	Classification: Covid-19, Model Evaluation	97.82% (Accuracy)	A proposal: best deep learning network
Ezz et al. [15]	VGG19, DenseNet121, ResNetV2, InceptionV3, InceptionResNetV2, Xception, and MobileNetV2	Classification: Covid-19, Model Evaluation	F1-scores : Normal :0.91 Covid-19 : 0.89	A proposal: best deep learning network
Salman et al. [43]	InceptionV3	Classification: Covid-19	100% (specificity) 100% (accuracy) 100% ( PPV) 100%, (NPV) 100% (f-1 score)	A proposed model, implementation and evaluation
Ali et al. [33]	ResNet50, InceptionV3 and Inception-ResNetV2	Classification: Covid-19	98% (accuracy) 96% (recall) and 100%(specificity)	A proposal: best deep learning network

TABLE V: Summary of papers that used Deep Learning in the field of COVID-19

Iv Future Research Opportunities

In this section, we have briefly presented the challenges and further research opportunities on AI/ML not only to fight against COVID-19 pandemic but also for the future pandemic.

Iv-a Importance of study context of future research

We observed that only one-third of the research (37%) used contextual data while the rest (63%) of them conducted research using data from more than one country. The countries or regions that are affected more than others have more opportunities to conduct contextual research. As China was the primary hotspot for the pandemic, a comparatively greater number of contextual studies have been conducted in China owing to the availability of more data and increased time to observe the nature of the pandemic. As the pandemic progressed, data from other countries also became available. Hence, there is considerably more scope for future contextual research that will aim to explore and predict the similarity of the pattern of the pandemic among Chinese studies and other regional studies.

Iv-B Potential areas of research

The existing research has been conducted to detect and diagnose COVID-19, epidemic forecasting, sustainable development, and patient management. We observe that a relatively small number (11.4%) of research has been conducted on epidemic forecasting, sustainable development and patient management. Further research can be done focusing on these areas. We observed that studies on epidemic forecasting and sustainable development used contextual data. We think that epidemic forecasting based research should always be contextual.

Iv-C Collecting and storing various types of data

There are opportunities to collect different types of data (e.g., images, texts, videos, etc.) and making it available for the researchers to conduct different experiments. Such efforts will be highly valuable for fighting the pandemic.

Iv-D Disease diagnosis and prediction research using a large set of data

Most disease detection and half of disease diagnosis-based research were conducted using global data. However, we suggest more research in this direction could use diverse global data in the future for better performance. A few research studies (24%) on disease detection and the other half research on disease diagnosis also performed context-based analysis on specific regions. Hence, future studies may consider other affected regions for disease pattern exploration as well. It can be insinuated from the studies considered in this review paper that a significant amount of data was not used for conducting machine learning and deep learning research. Future research could investigate if bigger datasets could result in better structured, authenticated and generalized outcomes. Additional work can be done to validate some of the original research studies ([26, 39]) that did not use data and only proposed models. The claims found in the studies can be explored more with sufficient data in the future.

Iv-E Exploring the effect and variation in research outcomes based on different types of data

The majority of the existing research that has developed an AI/ML-based tool with the purpose of disease detection and progression has employed training and testing methods. The methods, in general, require related datasets to train and validate the systems to correctly predict the outcome of a given problem, which in this case is detecting the disease accurately. Several deep learning based research (48%) studies have used chest X-Ray and CT scan images to determine the presence of opacity in lungs that depicts COVID-19. However, there are other lung diseases i.e., Pneumonia, COPD, Asthma etc. due to which similar effects on lungs X-Ray and CT images can be observed. Future studies can include other symptoms of patients in the format of text to train the system along with the images of lungs to determine COVID-19 from other diseases.

Iv-F Explore the contextual effect and variation in research outcomes

Only one cross-country research has been conducted with the purpose of sustainable development, where correlation among confirmed cases, environmental and demographic factors of four different countries were calculated and compared. Other cross-country research studies, similar to the one stated above, could be pursued in the future to determine if the virus spread depends on the environmental factors.

Iv-G Managing the ICU surge during the COVID-19 crisis

It has been reported that the hospitals decided to provide service only to the young people, leaving elder citizens who have less survival possibility as the hospitals were out of capacity. Further research can be pursued to predict which patient has higher likelihood to be a critical case based on their medical history and symptoms. This would help the hospitals determine which patient can be cured at home and who will need ICU support. Research studies focusing on ICU admission would be helpful on early releasing of some patients, and thereby, leaving space for the patients who need it most. At the same time, studies can also prevent early termination from ICUs.

Iv-H Supporting the health care workforce

The lack of health professionals has been observed in highly affected areas. They had to work beyond their limit making them vulnerable to human errors. AI-assisted systems could be helpful here. A rule-based AI can monitor all the data in the ICU and suggests the professionals to take necessary steps. An efficient AI can help allocate and control the flow of oxygen level: a crucial treatment given to the COVID-19 patients.

Iv-I Developing drugs/vaccine

Researchers all over the world currently have been working on developing drugs and vaccines for COVID-19. Different organizations have already been using AI to find a vaccine for COVID-19. From data analysis to decoy generation of COVID-19, AI can be helpful. Therefore, there are plenty of opportunities for improvement of these AI algorithms (Rosetta[23] and Quark[53]). Furthermore, AI can be used for simulation and analysis of different candidate vaccine.

V Conclusion

This paper presents a systematic review of exploring AI and ML techniques in fighting the COVID-19 pandemic. A total of 35 research articles were reviewed. These papers are analyzed and compared in various dimensions including the data types, input features, the AI techniques (the machine learning classification algorithms), as well as their objectives. Overall, there are three main contributions of our work. First, we have provided a summary of the findings in terms of objectives, data sources, data type, and volumes. Second, we have explored AI techniques that include various machine learning and deep neural network techniques in the field and compared their outcomes considering several popular evaluation metrics. Finally, we have identified several research issues based on our analysis and introduced corresponding new directions for future research.

Our study has a number of limitations, but at the same time, it provides some avenues for future research. First, we have used some specific keywords for searching the relevant materials. Although, our search keywords provided effective results to achieve the goal of our study, there might be a risk to miss some important materials that did not emerge from our search queries. Second, we think timely and up-to-date materials related to coronavirus and AI techniques are the key things that we have identified, studied, and summarized in this paper. Therefore, future works would be needed to collect more resources. Third, in addition to exploring AI and ML techniques in fighting the Covid-19 pandemic, future research can be conducted to analyze data privacy and security in the relevant areas. Although we have several limitations mentioned above, our analysis and discussion can have significant implications for both the health practitioners and researchers. We believe that our review opens a promising path and can be used as a reference guide for future research in this area.

References

[1] A. S. Ahuja (2019-10) The impact of artificial intelligence in medicine on the future role of the physician. PeerJ 7. External Links: Document Cited by: §I.
[2] A. Alimadadi, S. Aryal, I. Manandhar, P. B. Munroe, B. Joe, and X. Cheng (2020) Artificial intelligence and machine learning to fight covid-19. Physiological Genomics 52 (4), pp. 200–202. Cited by: §III-B, §III-D, TABLE II.
[3] Z. Allam and D. S. Jones (2020) On the coronavirus (covid-19) outbreak and the smart city network: universal data sharing standards coupled with artificial intelligence (ai) to benefit urban health monitoring and management. Healthcare 8 (1), pp. 46. Cited by: §III-B, TABLE II.
[4] I. D. Apostolopoulos and T. A. Mpesiana (2020-07) Covid-19: automatic detection from x-ray images utilizing transfer learning with convolutional neural networks. Physical and Engineering Sciences in Medicine. Cited by: TABLE I, TABLE IV, TABLE V.
[5] K. E. Asnaoui and Y. Chawki (2020) Using x-ray images and deep learning for automated detection of coronavirus disease. Journal of Biomolecular Structure and Dynamics 0 (0), pp. 1–12. Cited by: TABLE I.
[6] H. X. Bai, R. Wang, Z. Xiong, B. Hsieh, K. Chang, K. Halsey, T. M. L. Tran, J. W. Choi, D. Wang, L. Shi, J. Mei, X. Jiang, I. Pan, Q. Zeng, P. Hu, Y. Li, F. Fu, R. Y. Huang, R. Sebro, Q. Yu, M. K. Atalay, and W. Liao (2020) AI augmentation of radiologist performance in distinguishing covid-19 from pneumonia of other etiology on chest ct. Radiology, pp. 201491. Cited by: TABLE I, TABLE III.
[7] J. Bullock, A. Luccioni, K. H. Pham, C. S. N. Lam, and M. Luengo-Oroz (2020) Mapping the landscape of artificial intelligence applications against covid-19. arXiv. External Links: Link Cited by: §III-B, TABLE II.
[8] C. Butt, J. Gill, D. Chun, and B. A. Babu (2020-04) Deep learning system to screen coronavirus disease 2019 pneumonia. Applied Intelligence (). Cited by: TABLE I.
[9] F. de Dombal (1984) Computer-aided diagnosis of acute abdominal pain. the british experience. Revue d’epidemiologie et de sante publique 32 (1), pp. 50—56. External Links: ISSN 0398-7620 Cited by: §I.
[10] R. C. Deo (2015) Machine learning in medicine. Circulation 132 (20), pp. 1920–1930. External Links: Document Cited by: §I.
[11] S. E. Dilsizian and E. L. Siegel (2014) Artificial intelligence in medicine and cardiac imaging: harnessing big data and advanced computing to provide personalized medical diagnosis and treatment. Current Cardiology Reports 16 (1). Cited by: §I.
[12] B. Ghoshal and A. Tucker (2020) Estimating uncertainty and interpretability in deep learning for coronavirus (covid-19) detection. arXiv. External Links: Link Cited by: §III-B, §III-E, TABLE I, TABLE IV.
[13] O. Gozes, M. Frid-Adar, H. Greenspan, P. D. Browning, H. Zhang, W. Ji, A. Bernheim, and E. Siegel (2020) Rapid ai development cycle for the coronavirus (covid-19) pandemic: initial results for automated detection & patient monitoring using deep learning ct image analysis. arXiv. External Links: Link Cited by: §III-B, §III-C, §III-C, TABLE I, TABLE III, TABLE V.
[14] O. Gozes, M. Frid-Adar, H. Greenspan, P. D. Browning, H. Zhang, W. Ji, A. Bernheim, and E. Siegel (2020) Rapid ai development cycle for the coronavirus (covid-19) pandemic: initial results for automated detection & patient monitoring using deep learning ct image analysis. arXiv. External Links: Link Cited by: TABLE I, TABLE IV.
[15] E. E. Hemdan, M. A. Shouman, and M. E. Karar (2020) COVIDX-net: a framework of deep learning classifiers to diagnose covid-19 in x-ray images. arXiv. External Links: Link Cited by: §III-B, §III-E, TABLE I, TABLE IV, TABLE V.
[16] Z. Hu, Q. Ge, S. Li, L. Jin, and M. Xiong (2020) Artificial intelligence forecasting of covid-19 in china. arXiv. External Links: Link Cited by: §I, §III-B, §III-C, TABLE I, TABLE III, TABLE IV.
[17] F. Jiang, Y. Jiang, H. Zhi, Y. Dong, H. Li, S. Ma, Y. Wang, Q. Dong, H. Shen, and Y. Wang (2017-12) Artificial intelligence in healthcare: past, present and future. Stroke and Vascular Neurology 2 (4), pp. 230–243. Cited by: §I.
[18] X. Jiang, M. Coffee, A. Bari, J. Wang, X. Jiang, J. Huang, J. Shi, J. Dai, J. Cai, T. Zhang, Z. Wu, G. He, and Y. Huang (2020) Towards an artificial intelligence framework for data-driven prediction of coronavirus clinical severity. Computers, Materials & Continua 63 (1), pp. 537–551. Cited by: §III-B, §III-C, §III-E, TABLE I, TABLE III, TABLE IV.
[19] S. Johnstone (2020) A viral warning for change. the wuhan coronavirus versus the red cross: better solutions via blockchain and artificial intelligence. SSRN Electronic Journal. Cited by: §III-B, TABLE II.
[20] A. I. Khan, J. L. Shah, and M. M. Bhat (2020) CoroNet: a deep neural network for detection and diagnosis of covid-19 from chest x-ray images. Computer Methods and Programs in Biomedicine 196, pp. 105581. External Links: ISSN 0169-2607 Cited by: TABLE I.
[21] E. Kolker, V. Özdemir, and E. Kolker (2016-06) How healthcare can refocus on its super-customers (patients, n =1) and customers (doctors and nurses) by leveraging lessons from amazon, uber, and watson. OMICS: A Journal of Integrative Biology 20 (7), pp. 329–333. Cited by: §I.
[22] M. Kundu, M. Nasipuri, and D. K. Basu (2000) Knowledge-based ecg interpretation: a critical review. Pattern Recognition 33 (3), pp. 351–373. Cited by: §I.
[23] A. Leaver-Fay, M. Tyka, S. M. Lewis, O. F. Lange, J. Thompson, R. Jacak, K. W. Kaufman, P. D. Renfrew, C. A. Smith, W. Sheffler, et al. (2011) ROSETTA3: an object-oriented software suite for the simulation and design of macromolecules. In Methods in enzymology, Vol. 487, pp. 545–574. Cited by: §IV-I.
[24] L. Li, L. Qin, Z. Xu, Y. Yin, X. Wang, B. Kong, J. Bai, Y. Lu, Z. Fang, Q. Song, K. Cao, D. Liu, G. Wang, Q. Xu, X. Fang, S. Zhang, J. Xia, and J. Xia (2020-03) Artificial intelligence distinguishes covid-19 from community acquired pneumonia on chest ct. Radiology, pp. 200905. Cited by: §III-B, TABLE I, TABLE IV, TABLE V.
[25] M. Loey, F. Smarandache, and N. E. M. Khalifa (2020) Within the lack of chest covid-19 x-ray dataset: a novel detection model based on gan and deep transfer learning. Symmetry 12 (4), pp. 651. Cited by: §III-E, TABLE I, TABLE IV.
[26] H. S. Maghdid, K. Z. Ghafoor, A. S. Sadiq, K. Curran, D. B. Rawat, and K. Rabie (2020) A novel ai-enabled framework to diagnose coronavirus covid 19 using smartphone embedded sensors: design study. arXiv. External Links: Link Cited by: §III-D, TABLE I, §IV-D.
[27] B. McCall (2020-02) COVID-19 and artificial intelligence: protecting health-care workers and curbing the spread. The Lancet Digital health 2 (4), pp. e166–e167. Cited by: §III-B, TABLE II.
[28] X. Mei, H. Lee, K. Diao, M. Huang, B. Lin, C. Liu, Z. Xie, Y. Ma, P. M. Robson, M. Chung, A. Bernheim, V. Mani, C. Calcagno, K. Li, S. Li, H. Shan, J. Lv, T. Zhao, J. Xia, Q. Long, S. Steinberger, A. Jacobi, T. Deyer, M. Luksza, F. Liu, B. P. Little, Z. A. Fayad, and Y. Yang (2020-05) Artificial intelligence–enabled rapid diagnosis of patients with covid-19. Nature Medicine. Cited by: TABLE I, TABLE III.
[29] R. A. Miller, M. A. McNeil, S. M. Challinor, F. E. Masarie, and J. D. Myers (1986) The internist-1/quick medical reference project - status report. West J Med 145 (6), pp. 816–822. Cited by: §I.
[30] R. A. Miller (1994-01) Medical Diagnostic Decision Support Systems—Past, Present, And Future: A Threaded Bibliography and Brief Commentary. Journal of the American Medical Informatics Association 1 (1), pp. 8–27. Cited by: §I.
[31] T. B. Murdoch and A. S. Detsky (2013-04) The inevitable application of big data to health care. The Journal of the American Medical Association 309 (13), pp. 1351–1352. Cited by: §I.
[32] K. Murphy, H. Smits, A. J.G. Knoops, M. B.J.M. Korst, T. Samson, E. T. Scholten, S. Schalekamp, C. M. Schaefer-Prokop, R. H. H. M. Philipsen, A. Meijers, J. Melendez, B. van Ginneken, and M. Rutten (2020) COVID-19 on the chest radiograph: a multi-reader evaluation of an ai system. Radiology 0 (0), pp. 201874. Cited by: TABLE I.
[33] A. Narin, C. Kaya, and Z. Pamuk (2020-03) Automatic detection of coronavirus disease (covid-19) using x-ray images and deep convolutional neural networks. pp. . Cited by: TABLE I, TABLE IV, TABLE V.
[34] W. Naudé (2020) Artificial intelligence against covid-19: an early review. IZA Discussion Papers Technical Report 13110, Institute of Labor Economics (IZA). External Links: Link Cited by: §III-B, TABLE II.
[35] W. H. Organization (2020) Public health emergency of international concern (pheic). Cited by: §I.
[36] W. H. Organization (2020) WHO director-general’s opening remarks at the media briefing on covid-19 - 11 march 2020. External Links: Link Cited by: §I.
[37] U. Özkaya, Ş. Öztürk, and M. Barstuğan (2020-04)
Coronavirus (covid-19) classification using deep features fusion and ranking technique
. pp. . Cited by: §III-B, §III-C, TABLE I, TABLE III, TABLE IV.
[38] B. Pirouz, S. Shaffiee Haghshenas, S. Shaffiee Haghshenas, and P. Piro (2020) Investigating a serious challenge in the sustainable development process: analysis of confirmed cases of covid-19 (new type of coronavirus) through a binary classification using artificial intelligence and regression analysis. Sustainability 12 (6), pp. 2427. Cited by: §III-B, §III-C, §III-C, §III-D, §III-E, TABLE I, TABLE III, TABLE IV.
[39] S. Rahmatizadeh, S. Valizadeh-Haghi, and A. Dabbagh (2020-04) The role of artificial intelligence in management of critical covid-19 patients. Shahabedin Rahmatizadeh Saeideh Valizadeh-Haghi Ali Dabbagh 5 (1), pp. 16–22. Cited by: §III-B, TABLE I, §IV-D.
[40] A. S. R. S. Rao and J. A. Vazquez (2020) Identification of covid-19 can be quicker through artificial intelligence framework using a mobile phone-based survey when cities and towns are under quarantine. Infection control and hospital epidemiology 41 (7), pp. 826–830. Cited by: TABLE IV.
[41] A. S. R. S. Rao and J. A. Vazquez (2020-07) Identification of covid-19 can be quicker through artificial intelligence framework using a mobile phone–based survey when cities and towns are under quarantine. Infection Control & Hospital Epidemiology 41 (7), pp. 826–830. Cited by: §I, §III-D, TABLE I.
[42] S. Russell and P. Norwig (1995) Artificial intelligence: a modern approach. Neurocomputing 9 (2), pp. 215 – 218. Note: Control and Robotics, Part II External Links: ISSN 0925-2312, Document Cited by: §I.
[43] F. M. Salman, S. S. Abu-Naser, E. Alajrami, B. S. Abu-Nasserand, and B. A. M. Ashqar (2020-03) COVID-19 detection using artificial intelligence. International Journal of Academic Engineering Research (IJAER) 4 (3), pp. 18–25. Cited by: §III-D, TABLE IV, TABLE V.
[44] K. C. Santosh (2020-03) AI-driven tools for coronavirus outbreak: need of active learning and cross-population train/test models on multitudinal/multimodal data. Journal of Medical Systems 44 (5), pp. 93. Cited by: §III-B, TABLE II.
[45] I. H. Sarker, A. Kayes, and P. Watters (2019) Effectiveness analysis of machine learning classification models for predicting personalized context-aware smartphone usage. Journal of Big Data 6 (1), pp. 57. Cited by: §I.
[46] I. H. Sarker and F. D. Salim (2018) Mining user behavioral rules from smartphone data through association analysis. In Pacific-Asia Conference on Knowledge Discovery and Data Mining, pp. 450–461. Cited by: §I.
[47] P. K. Sethy and S. K. Behera (2020) Detection of coronavirus disease (covid-19) based on deep features. Preprints 2020030300, pp. 2020. Cited by: TABLE IV, TABLE V.
[48] F. Shi, J. Wang, J. Shi, Z. Wu, Q. Wang, Z. Tang, K. He, Y. Shi, and D. Shen (2020) Review of artificial intelligence techniques in imaging data acquisition, segmentation and diagnosis for covid-19. IEEE Reviews in Biomedical Engineering (), pp. 1–1. Cited by: §II, §III-B, TABLE II.
[49] P. Szolovits, R. S. Patil, and W. B. Schwartz (1988) Artificial intelligence in medical diagnosis. Annals of Internal Medicine 108 (1), pp. 80–87. External Links: Document Cited by: §I.
[50] L. Wang, Z. Q. Lin, and A. Wong (2020) COVID-net: a tailored deep convolutional neural network design for detection of covid-19 cases from chest x-ray images. arXiv. External Links: Link Cited by: §III-B, §III-E, TABLE I.
[51] S. Wang, B. Kang, J. Ma, X. Zeng, M. Xiao, J. Guo, M. Cai, J. Yang, Y. Li, X. Meng, and B. Xu (2020) A deep learning algorithm using ct images to screen for corona virus disease (covid-19). medRxiv. External Links: Link Cited by: §III-E, TABLE III, TABLE V.
[52] S. Wang, B. Kang, J. Ma, X. Zeng, M. Xiao, J. Guo, M. Cai, J. Yang, Y. Li, X. Meng, and B. Xu (2020) A deep learning algorithm using ct images to screen for corona virus disease (covid-19). medRxiv. External Links: Document, Link, https://www.medrxiv.org/content/early/2020/04/24/2020.02.14.20023028.full.pdf Cited by: §I, §III-C, TABLE I, TABLE IV.
[53] D. Xu and Y. Zhang (2012) Ab initio protein structure assembly using continuous structure fragments and optimized knowledge-based force field. Proteins: Structure, Function, and Bioinformatics 80 (7), pp. 1715–1735. Cited by: §IV-I.
[54] L. Yan, H. Zhang, J. Goncalves, Y. Xiao, M. Wang, Y. Guo, C. Sun, X. Tang, L. Jin, M. Zhang, X. Huang, Y. Xiao, H. Cao, Y. Chen, T. Ren, F. Wang, Y. Xiao, S. Huang, X. Tan, N. Huang, B. Jiao, Y. Zhang, A. Luo, L. Mombaerts, J. Jin, Z. Cao, S. Li, H. Xu, and Y. Yuan (2020) A machine learning-based model for survival prediction in patients with severe covid-19 infection. medRxiv. External Links: Link Cited by: §III-B, §III-C, TABLE I, TABLE III, TABLE IV.
[55] Z. Yang, Z. Zeng, K. Wang, S. Wong, W. Liang, M. Zanin, P. Liu, X. Cao, Z. Gao, Z. Mai, J. Liang, X. Liu, S. Li, Y. Li, F. Ye, W. Guan, Y. Yang, F. Li, S. Luo, Y. Xie, B. Liu, Z. Wang, S. Zhang, Y. Wang, N. Zhong, and J. He (2020) Modified seir and ai prediction of the epidemics trend of covid-19 in china under public health interventions. Journal of Thoracic Disease 12 (3). Cited by: §III-B, §III-C, §III-D, §III-E, TABLE I, TABLE III, TABLE IV.
[56] K. H. Yu, A. L. Beam, and I. S. Kohane (2018) Artificial intelligence in healthcare. Nature Biomedical Engineering 2 (10), pp. 719–731. Cited by: §I.

A Survey on the Use of AI and ML for Fighting the COVID-19 Pandemic

Authors

Mapping the Landscape of Artificial Intelligence Applications against COVID-19

Diagnosis of COVID-19 Using Machine Learning and Deep Learning: A review

Machine Learning based Anomaly Detection for Smart Shirt: A Systematic Review

Did Chatbots Miss Their 'Apollo Moment'? A Survey of the Potential, Gaps and Lessons from Using Collaboration Assistants During COVID-19

Can Machine Learning Catch the COVID-19 Recession?

Spike2Vec: An Efficient and Scalable Embedding Approach for COVID-19 Spike Sequences

Considerations, Good Practices, Risks and Pitfalls in Developing AI Solutions Against COVID-19

I Introduction

Ii Methodology