Artificial Intelligence in COVID-19 Management: A Systematic Review

: With the development of modern technologies in the field of healthcare, the use of Artificial Intelligence (AI) in disease management is increasing. AI methods may assist healthcare providers in the COVID-19 era. The current study aimed to observe the efficacy and importance of AI for managing the COVID-19 pandemic. An organized search was conducted, utilizing PubMed, Web of Science, Scopus, Embase, and Cochrane up to September 2022. Studies were considered qualified for inclusion if they met the inclusion criterion. We conducted review according to the Preferred Reporting Items for Systematic reviews and Meta Analyses (PRISMA) guidelines. There were 52 documents that met the eligibility criteria to be included in the review. The most common item using AI during the COVID-19 era was predictive models to foretell pneumonia and mortality risks in people with COVID-19 based on medical and experimental parameters. COVID-19 mortality was related to being male and elderly based on the Artificial Neural Network (ANN) and Convolutional Neural Network (CNN) logistic regression analysis of demographics,


Introduction
SARS-CoV-2 the cause of coronavirus disease  primarily emerged in Wuhan, China, in December 2019 and the World Health Organization (WHO) acknowledged a COVID-19 worldwide disease on March 11 th , 2020 (WHO, 2020a;Mehraeen et al., 2021).The severity of the disease ranges from flu like symptoms including fever, fatigue, cough, headache, diarrhea, myalgia, and sore throat, to atypical pneumonia causing Acute Respiratory Distress Syndrome (ARDS) with dyspnea, loss of consciousness, and chest pain (WHO, 2020b).According to the latest WHO reports, as of August 2 nd , 2022, this ongoing catastrophic pandemic has infected 575,887,049 cases and led to 6,398,412 deaths (WHO, 2022).COVID-19's long asymptomatic incubation period, relatively high reproduction numbers, and high mortality rates mostly among vulnerable patients (e.g., >65 years, immunocompromised, morbidities) have put an unprecedented burden on healthcare organizations everywhere.The combat against COVID-19 seemed an arduous task since the beginning due to overwhelmed hospitals, exhausted healthcare providers, medical supplies shortage, and detection tool kits (real time polymerase chain reaction) (Dadras et al., 2022).To control the pandemic and halt the rapid spread of the disease, many vaccines were introduced and granted emergent safety approvals by the Food and Drug Administration (FDA) and WHO.As of July 26 th , 2022,  12,248,795,623 vaccine doses have been ordered globally.However, despite the efficacy of they are not yet a definitive solution because of vaccine inequality, vaccine hesitancy, and new variants of the virus (WHO, 2022;Oliaei et al., 2021).
Thus, all these barriers signify the importance of new technologic methods in controlling the pandemic.For this reason, the use of Artificial Intelligence (AI) and Machine Learning (ML) has gained great popularity in different health systems globally over the past two decades.This has occurred due to the easy accessibility of data, the ubiquity of computers, and increasing computational power.Thus, AI and ML-based solutions have the exceptional capability in addressing the aforementioned issues (Shamsabadi et al., 2022;Hamet and Tremblay, 2017;Mehraeen et al., 2022).
AI and ML can be used in the diagnosis of COVID-19 through image processing and analysis of X-rays, CT scans, and ultrasounds.For instance, these methods can be used to differentiate between COVID-19 and other causes of pneumonia (Ulhaq et al., 2020).In addition, AI-based methods are used in COVID-19 control and prevention; deep learning models have been used to recognize mask wearing, infrared thermography techniques were utilized for fever detection (Somboonkaew et al., 2017) and mobile based applications were available for self-claimed COVID-19 symptomatic patients (Lahiri et al., 2012).Additionally, AI has been used in the clinical management of COVID-19 by selecting the most efficient treatment based on the severity of the disease and the patient's clinical condition (Siam et al., 2020).Finally, AI-based technique has also been used in the COVID-19 vaccine and medication development to find the most efficient lead components and chemical substances (Tang et al., 2022).
Recently, AI technologies such as ML-based prototypes trained on specific biomolecules have provided low cost and fast implementation approaches for the detection of practical viral treatments.However, there are not many articles on the application of this technology for pandemic management and so we aimed to investigate AI and ML's use, efficacy, and importance amid the COVID-19 pandemic and find the key differences between various ML models.

Materials and Methods
This study is an organized review of current literature pertinent to AI-based detection of COVID-19 disease.We have studied papers available in the English language as of September 2021.With the purpose of reliability and authenticity of the outcomes, this investigation adheres to the Preferred Reporting Items for Systematic reviews and Meta Analyses (PRISMA) checklist (Moher et al., 2009).

Data Sources
A search from December 2019 to September 2022 was directed using the following databases: PubMed, Web of Science, Scopus, Embase, and Cochrane.The search strategy employed combining the terms: "COVID-19" OR "SARS-CoV-2" OR "Coronavirus" AND "Artificial Intelligence (AI)" OR "deep learning" OR "machine learning" OR "data mining" OR "artificial neural networks" OR "deep neural networks" OR "convolutional neural networks" AND "detection" OR "diagnosis" OR "prognoses" OR "prognosis" OR "assessment" OR "distinction" OR "recognition".Searches were limited to documents available in the English language.Titles and abstracts of recovered articles were individually evaluated by five authors to assess their eligibility for review.Any disagreements unable to be solved following discussion were adjudicated amongst the authors.When abstracts did not provide sufficient information to examine study eligibility, the full text was retrieved for evaluation.Subsequently, each study selected in the previous stage was fully evaluated and selected by four reviewers.

Eligibility Criteria
Studies were suitable for inclusion if they met the following measures: (1) Documents published in English; (2) Human studies, original articles, and papers with the experimental data.Studies were excluded if they met the following criteria: (1) Reviews, non-original editorials, and meta analyses; (2) Literature without available full texts, abstract papers, or conference abstracts; (3) Literature with doubts about duplication and/or reliability of results; and (4) Clinical Trials which were in progress without published outcomes and (5) Studies that did not explain the implemented AI-model.

Data Extraction
Four members of our research team individually assessed the full text documents and accompanied data extraction, using a regular template/spreadsheet. Data extracted included first author (reference) ID, type of study, country of study, target population, type of AI program, the purpose of using AI, type of data used, model and type of AI technique used, a sample size of training, classification measures and other information related to the aims of this review.To eliminate possible repetitions and/or crossovers, the selected publications and extracted data were checked by other researchers.

Results
The database search achieved 617 qualified studies and following the screening 52 full text documents met the inclusion standards and included in the final evaluation (Fig. 1).
Various predictive models to predict mortality (Das et al., 2020;Ünlü and Namlı, 2020;Booth et al., 2021;Gao et al., 2020;Karthikeyan et al., 2021;Marcos et al., 2021;Pan et al., 2020;Stachel et al., 2021) based on clinical and laboratory parameters of confirmed COVID-19 patients were the most used technologies of AI in COVID-19 pandemic management.Other studies predicted a 30-day mortality risk in patients with COVID-19 pneumonia (Halasz et al., 2021) and patients' chances of surviving a SARS-CoV-2 infection (Ikemura et al., 2021).For instance, there was a correlation between COVID-19 mortality and being male and elderly in the Artificial Neural Network (ANN) and Convolutional Neural Network (CNN) logistic regression analysis of demographics, clinical data, and laboratory tests of hospitalized COVID-19 patients (Lin et al., 2021).
Building a predictive model as a screening tool to identify people and areas with a higher risk of SARS-CoV-2 infection to be prioritized for testing (Booth et al., 2021;Dantas et al., 2021;Singh et al., 2022), early detection of COVID-19 (Siddiqui et al., 2021) and prediction system for discharged patients based on Computer Tomography (CT) scan images, (Shah et al., 2022) was reported by several studies.
In a related article, this vital finding mentioned that data mining could be used as a model to predict the side effects of COVID-19 (Yang et al., 2021).Another study reported the Odds Ratio (OR) and a data mining algorithm to investigate the risks of cardiac adverse events associated with the possible pharmacotherapies for COVID-19 outpatients (Yuan et al., 2021).Deep Neural Network and Convolutional Neural Network (CNN) models were used to detect coronavirus disease from CT Scan images (Turkoglu, 2021).We also identified that the presence of several techniques was used (logistic regression, Support Vector Machine (SVM), K-nearest neighbor, random forest, and gradient boosting) to diagnose and predict mortality among confirmed COVID-19 patients (Schöning et al., 2021;Das et al., 2020;Ünlü and Namlı, 2020;Singh et al., 2022;Abdulkareem et al., 2021;Zhang et al., 2020).

Fig. 2:
The frequency of AI using purpose for the management of COVID-19

Discussion
The main objective of this study was to consider AI and ML's use, efficacy, and importance amid the COVID-19 pandemic and find the main variances among various ML models.Our results demonstrated that AI methods such as data mining, machine learning, deep learning, logistic regression, support vector machine, neural networks, K-nearest neighbor, random forest, and gradient boosting could help manage COVID-19.A similar article (Ohno et al., 2022) reported that ML-based CT texture analysis is equally or more useful for predicting the time until CT for favipiravir treatment on COVID-19 patients than CT disease severity score (Ohno et al., 2022).Also, Liang et al. (2022) in a related article concluded that a new AI system based on deep learning and federated learning has high reliability in diagnosing COVID-19 based on CT, with or without clinical data (Liang et al., 2022).Finally, existing literature on the use of AI during the COVID-19 epidemic determines the benefits of AI use in the pandemic such as early diagnosis, predictions, and even though modeling of treatments.
Discussing the type of program and the purpose of each study simultaneously provides a helpful understanding of the setting of each study.Many studies shared the same frameworks, like using AI to diagnose COVID-19 patients, but they applied different methods such as deep learning, data mining, machine learning, logistic regression, and support vector machines on targeted populations.But simply said, the diagnosis and prognosis of COVID-19 were the global aims of these studies.Interestingly, 11 studies used models to predict the prognosis of COVID-19 patients.This was the most abundant framework, followed by models diagnosing COVID-19, which was the setting of 9 studies.Mix methods of AI were also used in the management of COVID-19, such as using a model to develop an app to diagnose or assess the prognosis of patients.The outstanding results of each framework are discussed in detail below.
By using laboratory markers or chest radiograph imaging, researchers provided their models with data necessary for diagnosing COVID-19 regardless of patients' history, manifestations, and physical exam results.Applying routine laboratory test results as data, (Baktash et al., 2021) established a ML model to detect asymptomatic individuals infected with COVID-19.The accuracy of their model covered a range of 74.48% up to 81.79% depending on the technique and algorithm (Baktash et al., 2021).By comparation of people's signs and the results of traditional COVID tests Machine learning algorithms and models can predict COVID-19 infection.Populations, where access to testing is limited, can be examined by these diagnostic methods.During the COVID-19 pandemic mobile health apps that monitor patients, by gathering signs such as persistent coughing, fever, fatigue, and anosmia in daily reports on their health status, can predict COVID-19 infection.Development of a mobile application for self-management and selfmonitoring among patients with COVID-19 allows data gathered to be used to forecast severe COVID-19 patients by ML models (Mohammad et al., 2021).ML algorithms allow identifying of COVID-19 patients.This method of AI is a tendency towards the application of innovative statistical approaches to defining results as a function of inputs.For example, (Cabitza et al., 2021) established compound ML models using data retrieved from 21 to 34 blood test results of 1624 patients reaching precisions of 75-78% to differentiate those infected with COVID-19 from those who were not (Cabitza et al., 2021).
Image processing and modeling for prediction were the two common methods of AI for the management of the pandemic.Clinical image processing is the basis of many diagnostic models, such as chest X-rays and chest CT scans that play a major role in diagnosing respiratory infections, especially COVID-19.AI image processing and interpretation algorithms can detect/recognize, assess, and classify COVID-19 by segmenting, detecting, and quantifying the images' suspicious regions.Segmentation, localization, pattern classification, and extraction of Regions of Interest (RoIs) of chest X-rays or CT images play a particular role in Image classification (Kaheel et al., 2021).Outstanding results from different countries show that using image processing to analyze lung X-ray images, COVID-19 cases could be identified among pneumonia and healthy controls (Irmak, 2020;Alsaade et al., 2021;Civit-Masot et al., 2020;Fontanellaz et al., 2021;SeyedAlinaghi et al., 2022).Yang et al. (2021) designed a framework to find out the best architecture, pre-processing and training parameters by pre-trained Convolutional Neural Network (CNN) models and using deep learning techniques for the COVID-19 CT-scan classification tasks.The accuracy score was above 96% in the diagnosis of COVID-19 using CT-scan images that confirm the results (Yang et al., 2021).
Same as diagnosis, by predicting the diagnosis of COVID-19 patients, we require clinical data, upon which physicians provide the patient with less or more intensive care.Due to the characteristics of SARS-CoV-2 infection, to predict the outcome, we could focus on respiratory signs and symptoms.Bolouran et al., designed a model which was able to predict the 48 h respiratory failure of COVID-19 patients, using 10 parameters including oxygen delivery mode, ESI value, gender, and race (Bolourani et al., 2021).Another diagnostic model designed in Italy predicted 30-day mortality based on clinical data as well as medical history and demographics.This model showed high sensitivity (94%) but had low specificity (37%) (Halasz et al., 2021).Vital signs have also been involved in this process which includes: Systolic blood pressure, respiratory rate, and pulse oximetry level, as well as other laboratory test results.The result is a prognostic model that predicts patients' survival chances.In this method, by comparing the vital signs of a sick person with the vital signs of a healthy person, taking into account age and gender, the survival chances of COVID-19 patients are predicted (Ikemura et al., 2021).Ivano Lodato et al. (2022) developed a ML model to predict both the mortality and severity associated with COVID-19 based on data gathered from medical records and test results collected during their hospitalization.Decision tree, random forest, gradient, and RUS Boosting models of ML were used to test the accuracy of these models.Their results showed that random forest and gradient boosting classifiers were highly accurate in predicting patients' mortality (average accuracy ∼of 99%) (Lodato et al., 2022).COVID-19 computer model using the biochemical markers, inflammatory biomarkers and a Complete Blood Count (CBC) was another method mentioned in most of the studies included in this review.This model helps the physicians form an idea about the patient's overall status (Domínguez-Olmedo et al., 2021;Karthikeyan et al., 2021;Aktar et al., 2021).
Biochemical markers, such as Arterial Blood Gases (ABG), including pH, HCO3, O2, and CO2, are useful indicators of hemoglobin saturation status and are great importance in COVID-19.Using these in combination with inflammatory markers and CBC results along with some demographics, (Arvind et al., 2021) developed a model skilled at predicting the COVID-19 patients' necessity for intubation (Arvind et al., 2021).The unquestionable role of inflammatory biomarkers, during the course of COVID-19 made them one of the data targets for AI models and systems in COVID-19.Mimicking the follow-up protocols, some studies used inflammatory biomarkers as predictors of patients' outcomes.Levels of Lactate Dehydrogenase (LDH) and high-sensitivity C-Reactive Protein (hs-CRP) as useful indicators of a patient's inflammatory status helped with developing a model that predicted COVID-19 mortality with 90% accuracy 16 days before the outcome (Karthikeyan et al., 2021).Other laboratory values have also been integrated into AI models and systems.Some examples of these other laboratory markers include levels of D-dimer, troponin (Ikemura et al., 2021), and interleukin 6 (Chen et al., 2021).
Plain chest X-rays and chest CT scans are well-known diagnostic tools for COVID-19 and many other respiratory conditions and infections.Apart from COVID-19, interdisciplinary researchers have aimed to develop systems with the ability to interpret medical imaging modalities.Identifying chest radiographs or CT-scans that belong to known COVID-19 cases, while healthy and non-COVID-19 pneumonia cases were used as controls, describes the majority of study frameworks in this field (Ghaderzadeh et al., 2021;Irmak, 2020;Hwang et al., 2020;Khan, 2021;Xu et al., 2020;Sheikhbahaei et al., 2022;Behnoush et al., 2022).Age, demographics, chronic medical condition (Arvind et al., 2021), vital signs, exposures, and even gender were extracted from medical records and used to make the artificial models more realistic.In addition, novel approaches to diagnosis gathered attention among scientists.For instance, the system designed by Andreu-Perez et al. (2021) uses cough sounds in combination with quantitative RT-PCR and lymphocyte count to diagnose individuals infected with COVID-19 (Andreu-Perez et al., 2021).
One of the limitations of the current research was the breadth of methods and sub-branches of AI used in clinical care, so researchers had to study all the included articles in more detail and extract data in order to complete the table of results.Also, as interdisciplinary works, the included studies in this review were designed and conducted by researchers from different branches of science, mainly medicine, and computer sciences.Therefore, the interpretation of their results would have best been done through an interdisciplinary exchange of views.However, due to the specific aim of this review, it proceeded mostly from a medical point of view.

Conclusion
Managing difficult conditions in human life requires advanced technologies.COVID-19 is one of the important challenges in the health field that has involved the whole world.Information and communication technology tools such as AI can help manage this pandemic.In this research, the applications of artificial intelligence for managing COVID-19 were investigated and it was stated that AI can predict, diagnose and model COVID-19 by using techniques such as support vector machine, decision tree, and neural network.It is suggested that future research should deal with the design and development of AI-based tools for the management of chronic diseases such as COVID-19.

Table 1 :
Description of the findings reported in the eligible studies