Survey on Artificial Intelligence in the Management of COVID-19

Introduction

A global outbreak induced by SARS-CoV-2 (severe acute respiratory syndrome-CoronaVirus-2) caused COVID-19 pandemic, has spread rapidly around the world. Many governments, businesses and academic research organizations are contributing in the COVID-19 fight to stop the spread of the pandemic. Since December 2019, the COVID-19 was first reported in China, and then in various countries in the world. The World Health Organization (WHO) declared the outbreak as a Public Health Emergency of International Concern (PHEIC), and on March 2020, COVID-19 was recognized as a pandemic. COVID-19 resulted in the deaths of millions of people and the economic damage it imposed to the world was immense (Jianguo Chen, 2020, p. 1). Increased inflammatory mediators caused by COVID-19 lead to multiple organ failure, which increases the mortality risk in patients with severe disease (Hao Lv, 2021, p. 1). People have been negatively impacted socially, mentally, and economically as a result of pandemic crisis. It was believed that the outbreak was started by the zoonotic spread from the sea food markets in Wuhan, China. Later, it was thought that human-to-human transmission was to blame for the infection’s local spread over the globe, where approximately, 200 nations have been affected due to the pandemic. All people are susceptible to COVID-19 infection, although people aged 60 years and above and those with related symptoms are more likely to experience severe symptoms (Anjan Gudigar, November 2021).

Extensive research has been done on the virology, ecology, receptor interaction, origin and classification, cell entry, and physiological features of SARS-CoV-2. The major procedures used today for the disease detection and diagnosis includes medical imaging (such as chest X-ray or CT imaging), nucleic acid testing and serological diagnostics (Jianguo Chen, 2020, p. 1). When discussing the pathogenesis of SARS-CoV-2, virus introduction and spread, immune response and pathological signs are among the aspects that are highlighted. Numerous studies have been done in epidemiology on the origin and range of infections, epidemiological traits, clinical trials,  epidemic prediction, and tracking of transmission routes. Intensive care, vaccine development and drug development are potential therapies for COVID-19. The primary social control techniques now used are social isolation and communication prediction (Jianguo Chen, 2020, p. 2).

Artificial intelligence(AI) is a term that is used to describe a range of technologies including Machine Learning (ML), computer vision, knowledge graphs, biometrics, natural language processing, human-computer interaction (HCI), virtual reality, and augmented reality,  that enable computers to process information in a manner similar to that of human beings (Jianguo Chen, 2020). Artificial intelligence (AI) technologies are a powerful tool against COVID-19, widely used in fighting the pandemic. AI has previously been validated in a number of scientific domains, which has motivated academics to continue working on the COVID-19 issue. The COVID-19 outbreak has been significantly reduced because of the use of artificial intelligence techniques in a number of medical imaging modalities, including X-ray, computed tomography (CT), and ultrasound(US) (Anjan Gudigar, November 2021).

 

Literature Review

In the health care system, AI has been used for treatments, clinical decision making, diagnosis, and public health. AI techniques are crucial for the rapid detection of  COVID-19 patients during the current pandemic. In 2020, the number of studies utilizing AI methods, rapidly expanded. Many evaluations discuss the diagnosis of COVID-19 using chest CT-images and AI (Jianguo Chen, 2020).

Fortunately, research in the industry, medical and scientific sectors has successfully employed innovative AI technologies in the COVID-19 struggle and has made substantial progress in the short time since the emergence of COVID-19. AI assists in the diagnosis of COVID-19 by analyzing the medical images and offers non-invasive detection methods to stop the spread of infections to medical personnel. In virology research, artificial intelligence is employed to examine the structure of SARS-CoV-2 related proteins and predict novel compounds that could be exploited in the development of drugs and vaccines. Additionally, AI builds epidemic transmission models using social media and large-scale COVID-19 case data to precisely anticipate the outbreak time, route of transmission, range of transmission, and impact of disease. AI is also frequently employed in social control and epidemic prevention, such as in patient trajectory tracking, airport security checks, and epidemic visualization (Jianguo Chen, 2020, p. 2).

Medical image inspection is a widely used clinical method for identification and diagnosis of COVID-19. AI technology plays a crucial part in the medical image inspection and has made substantial advancements in organ recognition, image acquisition, disease classification and infection region segmentation. It not only significantly reduces radiologists’ time spent diagnosing patients through imaging, but also increases the accuracy of the diagnosis (Jianguo Chen, 2020, p. 4). Another use of AI is the contact tracking of COVID-19 patients. It is possible to discover potential transmission routes for each COVID-19 confirmed patient by integrating personal data (such as travel records, mobile phone positioning data and consumption records). Furthermore, mobile phone placement and AI frameworks can help the government better comprehend people’s status when they are experiencing social isolation (Jianguo Chen, 2020).

Following the global COVID-19 pandemic, the challenge was to transform the research results into novel technologies and drugs. Because of this, the demand for innovative drugs and rapidly development times has gained prominence in research. The frequent drug approval process has been significantly accelerated by the rise of AI. Early diagnosis of COVID-19 might be beneficial to stop the spread of the pandemic by facilitating quick decision making (Anjan Gudigar, November 2021, p. 1).

Objective

This study gives a brief view of the contributions and developments in the detection and diagnosis of COVID-19 using artificial intelligence techniques.

 

Methodology

RT-PCR Detection:

Real-time reverse transcriptase polymerase chain reaction (RT-PCR) is the current standard detection tool in the diagnosis of SARS-CoV-2 virus and bacterial infections, and it possess the benefits of high specificity and sensitivity. In early January 2020, the SARS-CoV-2 virus spread in the neighborhoods of Wuhan, China, according to 9 RNA positives that were found from pharyngeal swabs of patients using RT-PCR. Multiple routes of transmission are possible, according to the shedding of SARS-CoV-2 virus in lungs, feces, and throat. RT-PCR is constrained by difficult sample preparation, poor detection effectiveness, and a high percentage of false positive results (Jianguo Chen, 2020, p. 4).

For rapid screening of SARS-CoV-2, isothermal nucleic acid amplification and blood testing techniques are also frequently used methods. In order to extract routine hematological and biochemical parameters from blood testing and to offer COVID-19 classification, a machine learning (ML) classification algorithm was applied. A total of 105 blood test results were gathered, of which 27 were positive samples from people who had been diagnosed with COVID-19. Negative samples were taken from people who had lung cancer, TB, and common pneumonia for comparison. Each sample has 49 feature variables, consisting of 25 biochemical characteristics and 24 standard hematological parameters. Then the RF algorithm (Random forest- a machine learning algorithm used in classification and regression problems) (source: https://www.javatpoint.com/machine-learning-random-forest-algorithm) was applied to the training data to learn and classify features. Then an Rf classifier was constructed based on the retrieved 11 essential feature variables, examined 253 samples of 169 patients with probable COVID-19, and achieved an accuracy of 96.97%. Even though RT-PCR and blood tests rarely directly include AI technology, the viral load and COVID-19 case data gathered using serve as significant data sources for the future AI based analysis (Jianguo Chen, 2020, p. 4).

 

Medical Image Inspection:

·     CT Image Inspection:

For the early diagnosis of COVID-19, CT imaging provides a crucial foundation. Following steps are often involved in the progress of AI based CT image inspection for COVID-19:

lung tissue feature extraction, candidate infection region detection, Region of Interest (ROI) segmentation, and COVID-19 classification.

The first step is AI based image inspection is the segmentation of lung organs and ROIs. For further analysis and quantification, it shows the ROIs in lung CT scans (i.e., lungs. Lung lobes, bronchopulmonary segments, and infected regions or lesions). For CT image segmentation, many Deep learning (DL) models (such as, U-Net, V-Net, and VB-Net) have been employed. An enhanced segmentational model (named VB-Net) was proposed based on the V-Net and ResNet models using CT data from patients with confirmed COVID-19. In order to extract the ROIs from each CT image and identify the training curve of suspicious lesions, a DL model was created based on the U-Net++ framework. To separate the infection zones from the lung CT scans, a 3D DL model was applied. The segmented regional images were then split into 3 groups, such as, COVID-19, influenza-A viral pneumonia, and normal, using a classification model that was created using ResNet and location-attention structures. Each lung CT scan was used to extract the lung organs as ROIs using the U-Net segmentation model. From lung CT images, 18 lung areas and infected regions were precisely using the VB-Net model which was also utilized to produce 63 quantitative features (Jianguo Chen, 2020).

Different AI approaches were put out with a focus on the identification and location of candidate infection areas. Lung nodules and small opacity in the 3D lung volume were found using commercial software. A DL model was constructed which comprises of the ResNet and U-Net structures. In order to extract the ROI regions, the U-Net module was employed, and the ResNet model was used to detect and classify diffused turbidity and ground glass infiltration. Additionally, the CT scans of the patients with confirmed COVID-19 were compared with the patients without the virus, and thoroughly examined the CT features of COVID-19. Lung CT images were used to segment infected areas and organs, using a V-Net based CNN model. The best CT morphological features were then determined using the LASSO approach (Least Absolute Shrinkage and Selection Operator). Eventually, the severity of COVID-19 was anticipated and assessed based on the best CT morphology and clinical aspects. The collection of CT images included both negative and COVID-19 positive patients. By using the CNN model and the inception framework, COVID-19 disease can be predicted and randomly chosen ROI images were classified. The lung burden of individuals with COVID-19 was quantitatively assessed using an AI based InferReadTM CT pneumonia tool (Jianguo Chen, 2020, p. 5).

·     Chest X-ray Image Inspection:

Chest X-ray (CXR) images are simpler to obtain during radiological inspections than CT images. Despite being the common imaging technique for the diagnosis of COVID-19, CXR imaging is typically regarded as being less sensitive than CT imaging. Patients with early COVID-19 had normal features in some CXR images. Airspace opacity, GGO, and later mergers are radiological indicators of COVID-19 CXR imaging. Additionally, it is mainly noted how the lower, peripheral, and bilateral parts are distributed (Jianguo Chen, 2020).

Several research developed AI based classification models by nesting or fusing preexisting ML and DL models, with the focus on the categorization of COVID-19 based on CXR images. On the basis of CXR images, a DL framework (named COVIDX-Net) was developed to assist radiologists in autonomously diagnosing COVID-19 (Jianguo Chen, 2020, p. 8).

 

Drug Development:

Artificial intelligence technologies can be used in drug development industry to evaluate potential new treatments for COVID-19 by examining the interactions between current drugs and the COVID-19 protein targets. Additionally, by creating new molecular structures that inactivate proteases at molecular level, AI technologies can aid in the discovery of novel drug like compounds against COVID-19. To identify prospective COVID-19 drug candidates, a DL based molecular transformer-drug target interaction (MT-DTI) model was presented. To forecast the target proteins’ 3D crystal structures, the MT-DTI model uses SMILES string (Simplified molecular-input line-entry system) and amino acid sequences (Jianguo Chen, 2020).

 

Vaccine Development:

There are currently 3 different types of COVID-19 vaccine candidates available, such as, for the entire virus, for recombinant protein subunit and for nucleic acids. AI technology has contributed in the COVID-19 vaccines’ design and development. Contrary to the explicit uses of AI in other domains, the development of vaccines typically makes use of AI in an indirect way. The ability to predict immunological stimulation is a crucial component of vaccine design. Epitope and immunological interactions are typically predicted using various ML techniques and position specific scoring matrices (PSSM), which then predict the development of adaptive immunity in the target host. The COVID-19 vaccines for epitope prediction are developed using the AI methods of netMHC and netMHCpan. SARS-CoV-2 protein sequences were retrieved from GenBank and MSA technique was used to trim the nucleocapsid phosphoprotein sequences into potential peptide sequences. Based on this, the peptide sequences were being trained and predicted, using netMHC and netMHCpan AI systems (Jianguo Chen, 2020).

 

Conclusion

The COVID-19 struggle has successfully used AI technologies in particularly every field. In the above review we discussed the major contributions and developments of artificial intelligence technologies in different domains, i.e., in medical imaging (chest X-ray and CT imaging), drug development, vaccine development, and RT-PCR.

 

References

·         Anjan Gudigar, U. R. (November 2021). Role of Artificial Intelligence in COVID-19 Detection. sensors.

·         Hao Lv, L. S.-Y. (2021). Application of artificial intelligence and machine learning for COVID-19 drug discovery and vaccine design. Briefings in Bioinformatics, 1-10.

·         Jianguo Chen, K. L. (2020, july). A Survey on Applications of Artificial Intelligence in Fighting. Retrieved November 2022, from Cornell University

·         Kim, J. K. (2019, November). A Simple and Multiplex Loop-Mediated Isothermal Amplification (LAMP) Assay for Rapid Detection of SARS-CoV. Springer Link, 341-351.

Shigao Huang, J. Y. (2021). Artificial intelligence in the diagnosis of COVID-19. International Journal of Biological Sciences.