PaperlayerView exact revision on arXiv

Source-linked AI summary

Artificial Intelligence (AI) and Big Data for Coronavirus (COVID-19) Pandemic: A Survey on the State-of-the-Arts

Quoc-Viet Pham, Dinh C. Nguyen, Thien Huynh-The, Won-Joo Hwang, Pubudu N Pathirana

arXiv:2107.14040v1cs.CY

TL;DR

The COVID-19 pandemic created an urgent need for faster, data-informed responses across detection, prediction, treatment, and prevention. This paper surveys AI and big-data applications, representative frameworks, challenges, and recommendations. It concludes that these technologies can provide fast responses and meaningful information, while broader clinical implementation and better datasets remain needed.

  • Problem

    COVID-19 spread globally while clinical vaccines and specific treatments were unavailable, creating a need for effective detection, prediction, and response approaches.

  • Method

    The paper surveys AI and big-data applications, representative frameworks, associated challenges, and recommendations for pandemic response.

  • Results

    AI and big data support outbreak prediction and tracking, diagnosis and treatment, vaccine and drug discovery, and fast responses for medical staff and policymakers.

  • Takeaways & Limitations

    These technologies offer promising tools for understanding and mitigating COVID-19, but their value depends on continued research and support from official organizations.

  • Takeaways & Limitations

    AI studies surveyed were not implemented at large scale or tested clinically, and COVID-19 data quality and quantity require improvement.

Abstract

from arXiv · show

The very first infected novel coronavirus case (COVID-19) was found in Hubei, China in Dec. 2019. The COVID-19 pandemic has spread over 214 countries and areas in the world, and has significantly affected every aspect of our daily lives. At the time of writing this article, the numbers of infected cases and deaths still increase significantly and have no sign of a well-controlled situation, e.g., as of 13 July 2020, from a total number of around 13.1 million positive cases, 571, 527 deaths were reported in the world. Motivated by recent advances and applications of artificial intelligence (AI) and big data in various areas, this paper aims at emphasizing their importance in responding to the COVID-19 outbreak and preventing the severe effects of the COVID-19 pandemic. We firstly present an overview of AI and big data, then identify the applications aimed at fighting against COVID-19, next highlight challenges and issues associated with state-of-the-art solutions, and finally come up with recommendations for the communications to effectively control the COVID-19 situation. It is expected that this paper provides researchers and communities with new insights into the ways AI and big data improve the COVID-19 situation, and drives further studies in stopping the COVID-19 outbreak.

I. INTRODUCTION

COVID-19 rapidly became a global health and societal crisis without clinical vaccines or specific treatments. The paper surveys how AI and big data can support pandemic response, reviews applications and examples, and identifies challenges and recommendations.

  • COVID-19 spread rapidly worldwide and affected healthcare, education, transportation, politics, and supply chains.
  • As of July 13, 2020, more than 3.4 million confirmed cases and nearly 137,782 deaths were reported in the most affected country described.
  • The pandemic lacked clinical vaccines and specific drugs or therapeutic protocols, motivating alternative approaches for detection, prediction, and treatment.
  • Governments, international organizations, and technology companies responded through guidelines, information portals, supercomputing research, and open access to relevant literature.
  • The survey reviews AI and big-data applications, presents two representative frameworks, and discusses challenges and recommendations for researchers, governments, and societies.

B. ARTIFICIAL INTELLIGENCE

AI uses machine learning and deep learning to extract patterns and representations for intelligent applications, while big data supplies heterogeneous and rapidly generated information. In COVID-19 research, these technologies support drug discovery, diagnosis, prediction, and medical-data analysis.

  • Machine learning extracts meaningful patterns from data, whereas deep learning learns increasingly meaningful representations through multiple sequential layers.
  • Deep learning has been applied to identify existing drugs for rapid drug repurposing and to generate COVID-19 protease structures for molecular simulations.
  • A deep-learning model trained on 499 CT volumes and tested on 131 achieved accuracy 0.901, positive predictive value 0.840, and negative predictive value 0.982.
  • COVID-19 big data includes electronic healthcare records and information from physician notes, X-ray reports, case histories, devices, and outbreak areas.
  • Big data is characterized by volume, variety, and velocity, including text, images, videos, structured or unstructured data, and real-time updates.

2) Big data for COVID-19 fighting

Big data analytics supports COVID-19 response by extracting useful information from large datasets and combining with AI to model, monitor, and predict the pandemic. Its applications include outbreak tracking, understanding virus structure, treatment, and vaccine manufacturing.

  • Big data applications address outbreak tracking, virus structure, disease treatment, and vaccine manufacturing.
  • Combining big data with AI enables complex simulation models using coronavirus data streams for outbreak estimation.
  • These models support monitoring coronavirus spread and preparing preventive measures, while aggregated data supports future epidemic prediction.
  • Big data analytics collects and analyzes large datasets to discover hidden patterns and other information about COVID-19.

A. AI FOR COVID-19 DETECTION AND DIAGNOSIS

AI-based detection and diagnosis approaches address the need for faster, more accessible COVID-19 identification by analyzing smart-device data and medical images. The reviewed models also quantify disease severity and support outbreak prediction, while traditional SIR models require adaptation to COVID-19-specific dynamics.

  • Detection and diagnosis: RT-PCR testing is costly, time-consuming, equipment-dependent, and limited by testing-kit shortages, motivating low-cost smart-device and AI-based identification.These approaches are described as mobile health or mHealth solutions, with cloud and edge computing helping address device constraints.
  • Detection and diagnosis: Deep learning models use X-ray images and CT scans to automatically detect COVID-19 infection and support medical-image-based diagnosis.The paper identifies medical image processing as a major AI direction for COVID-19 detection.
  • Severity assessment: A random-forest model using 63 CT-derived features achieved sensitivity of 0.933, selectivity of 0.745, accuracy of 0.875, and AUC score of 0.91 for COVID-19 severity assessment.The study found severity was more dependent on features extracted from the right lung.
  • Outbreak prediction: Traditional SIR models assume recovered individuals cannot be reinfected and treat transmission and recovery rates as time-invariant, limiting their suitability for COVID-19.Time-dependent variants model β and γ as changing with interventions such as city lockdowns and traffic halts.
  • Outbreak prediction: Machine-learning and deep-learning models incorporate quarantine policies and transmission data to estimate outbreak size, including effects of social distancing and intervention timing.One modified autoencoder predicted outbreak size with an average error below 2.5%, while one-week intervention reduced peak cumulative and dead cases by around 166.89 times compared with one-month intervention.

C. AI FOR INFODEMIOLOGY AND INFOVEILLANCE

AI and big-data methods support infodemiology and infoveillance by analyzing accessible, timely information from social platforms and heterogeneous public-health, mobility, demographic, and user-generated sources. These analyses assess public responses and improve short-term COVID-19 case forecasting.

  • Data sources and public behavior: Online platforms provide highly accessible and timely COVID-19 information that can be analyzed when collected and processed properly.The paper identifies official health organizations and social channels as major sources of reliable pandemic information.
  • Data sources and public behavior: Analyses of Sina Weibo, Baidu, and Ali e-commerce data assess public concerns, risk perception, emotions, and behaviors during the COVID-19 outbreak.These sources support infodemiology and infoveillance studies of public responses in China.
  • Multisource outbreak monitoring: AI-based outbreak applications combine official health, demographic, mobility, and social-media data from heterogeneous sources.The reviewed systems integrate information across multiple levels rather than relying on a single data stream.
  • Multisource outbreak monitoring: Augmented ARGONet predicts confirmed COVID-19 cases two days ahead using health reports, Baidu searches, news activity, and model forecasts.Random Gaussian noise augments each data point before Lasso regression predicts cases for 32 Chinese provinces.
  • Multisource outbreak monitoring: Augmented ARGONet outperformed baseline models in most testing scenarios for predicting confirmed cases across 32 Chinese provinces.

D. AI FOR BIOMEDICINE AND PHARMACOTHERAPY

AI applications in biomedicine and pharmacotherapy use expanding COVID-19 data for drug discovery, diagnosis, and treatment support. The reviewed studies remain promising but face dataset and clinical-validation limitations.

  • AI and deep learning support biomedical research by analyzing large COVID-19 datasets for drug discovery and therapeutic development.The section links growing biomedical data with applications in biomedicine and pharmaceutical research.
  • A deep-learning model screened 4,895 commercially available drugs and identified ten potential inhibitors with high affinities.The model was trained on a COVID-19 virus-specific dataset.
  • CVL218 emerged as a candidate inhibitor and had a safety profile verified in monkeys and rats.The passage presents CVL218 as a potential COVID-19 treatment candidate, not a clinically approved medicine.
  • The section surveys AI applications spanning COVID-19 detection, diagnosis, outbreak tracking, infodemiology, infoveillance, biomedicine, and pharmacotherapy.
  • The reviewed AI studies provide fast responses and potentially meaningful information for medical staff and policymakers, despite limited large-scale or clinical testing.The authors emphasize that dataset quality and quantity require further improvement.

A. OUTBREAK PREDICTION

Big-data analytics support COVID-19 outbreak prediction by combining large-scale, geographically diverse data with pandemic models. These approaches estimate outbreak tendencies and support disease-control planning, while data completeness remains a concern.

  • Big-data analytics use large-scale COVID-19 datasets to estimate outbreak possibilities and inform disease-control planning.Italian Civil Protection data supported models of pandemic dynamics, while migration data supported quarantine-oriented prediction in Wuhan.
  • Authoritative Chinese health data enabled models and simulations of cumulative infections, recoveries, and outbreak tendencies across five regions.The regions were the Mainland, Hubei, Wuhan, Beijing, and Shanghai.
  • Outbreak-fitting accuracy may be uncertain when available data points do not capture comprehensive investigations or all relevant contributing factors.
  • Prediction studies used reports from China, South Korea, Italy, and Iran to generate daily infection estimates and potentially support longer-term outbreak forecasting.
  • Johns Hopkins data enabled short-term outbreak estimation in India, while U.S. city data supported prediction-error analysis for improving future models.The India trial used a two-week prediction interval.
  • B. VIRUS SPREAD TRACKING: Big-data solutions also track spread by combining national datasets and deriving models that estimate infection growth and regional maxima.One comprehensive model combined data from China, Singapore, South Korea, and Italy.
  • B. VIRUS SPREAD TRACKING: A 42-country analysis built a dataset covering 88 countries and reported lower northern-hemisphere growth rates associated with warmer weather and lockdown policies.
  • B. VIRUS SPREAD TRACKING: Because ground-truth surveillance data and model reliability are limited, an unsupervised approach used symptoms, news coverage, and transfer learning across countries.

C. CORONAVIRUS DIAGNOSIS/TREATMENT

Big-data methods support COVID-19 diagnosis and treatment through genomic, proteomic, structural, and clinical analyses. The reviewed work includes diagnostic assays, biological pathway analysis, immune-target prediction, and hospital-data guidelines.

  • Big-data applications support infectious-disease diagnosis, treatment-outcome prediction, and surgical decision-support tools.
  • A SARS-CoV-2 multiplex PCR scheme used 172 genome-specific primer pairs sourced from a national biological-information center.
  • Proteomic analysis examined 6,381 proteins from COVID-19-infected human cells using impact-pathway and network analyses.
  • A diagnosis procedure combined SARS-CoV-2 protein 3D-structure prediction with conformational B-cell-epitope prediction and epitope-conservation analysis.
  • A comprehensive guideline integrated epidemiology, prevention, diagnosis, and treatment, and analyzed screening data from 11,500 people at Wuhan University hospital.The passage states that 276 people were identified as suspected infectious cases.

D. VACCINE/DRUG DISCOVERY

Big-data methods support vaccine and drug discovery by mining sequence, epitope, and compound databases. The reviewed studies identify vaccine targets, predict immune responses, and screen existing drugs for inhibitory activity.

  • Big-data analysis of GISAID amino-acid sequences was used to seek potent targets for COVID-19 vaccine development.
  • Reverse vaccinology and immune informatics used NCBI entries and an epitope database to predict T-cell and B-cell epitopes.The workflow also considered antigenicity and allergenicity.
  • Molecular docking screened more than 2,500 FDA-approved small molecules, with 15 of 25 validated drugs showing significant inhibitory potencies.The findings supported drug repositioning against COVID-19.
  • Machine learning combined knowledge graphs and literature to support a big-data-driven drug-repositioning scheme for vaccine development.
  • The review identifies outbreak prediction, spread tracking, diagnosis and treatment, and vaccine or drug discovery as major big-data applications.

A. PHONE-BASED SOLUTION FOR DETECTION AND SURVEILLANCE

The paper presents two AI and big-data frameworks: mobile phones for COVID-19 detection and surveillance, and data-driven discovery of antibody sequences. It also identifies practical and clinical barriers to deploying these approaches.

  • Phone-based detection and surveillance: Mobile phones can support COVID-19 outbreak identification, diagnosis, treatment, case management, and disease elimination through wireless, sensing, and computing capabilities.The proposed framework uses cloud- or edge-trained deep learning models pushed to phones, which collect data through embedded cameras and biosensors.
  • Phone-based detection and surveillance: AI-based mobile health deployment is constrained by the capability and reliability of mobile hardware and software.The paper suggests standardized devices with dedicated computing and sensing components as one possible response.
  • Phone-based detection and surveillance: Diagnostic data collection and storage must address local policy requirements and patients’ trust, including preferences for face-to-face clinical interaction.Some policies may require results to remain on local servers rather than being transmitted to cloud servers used for globally trained models.
  • Phone-based detection and surveillance: Clinical cost and effectiveness remain difficult to evaluate because phone-based results may be questionable and not comparable with direct tests.This limits straightforward assessment of phone-based detection and surveillance frameworks.
  • Neutralizing antibody discovery: The antibody-discovery framework combines AI, big data, and medical knowledge to identify sequences that may inhibit COVID-19 virus growth.It starts from 1831 antigen and antibody sequences with corresponding IC50 values and uses molecular-graph features for model evaluation.
  • Neutralizing antibody discovery: XGBoost was selected as the strongest of five evaluated models, achieving 100% out-of-class prediction for SARS and Dengue, 84.61% for Influenza, and 75% for Ebola and Hepatitis.Using SARS-related antibody characteristics, the study generated 2589 hypothetical candidates and selected 8 potential neutralizing antibody structures.

VI. CHALLENGES, LESSONS, AND RECOMMENDATIONS

The paper highlights the need for coordinated policies and participation across authorities, residents, researchers, and industry as COVID-19 cases increase. Korea’s quarantine and mobile self-diagnosis policy illustrates one such response.

  • Recommendations: Regulatory authorities are positioned to define policies that encourage broad participation and harmonize approaches to controlling the outbreak.The relevant groups include residents, scientists, researchers, industry, technology companies, and large firms.
  • Lessons: COVID-19 control measures include lockdowns, social distancing, large-scale screening, and testing as confirmed cases rise.The paper frames these measures as approaches already being used to control the outbreak.
  • Lessons: From 1 April 2020, Korea required arriving passengers to quarantine for 14 days at registered addresses or designated facilities.The policy also required twice-daily self-diagnosis reporting through mobile applications.

2) Lack of standard datasets

The paper identifies nonstandardized datasets as a major obstacle to trustworthy COVID-19 AI and big-data applications. It discusses collaborative data collection and privacy-preserving or participation-enhancing approaches as potential responses.

  • Dataset standardization: Different datasets prevent reliable comparison of COVID-19 detection algorithms, even when reported accuracy, specificity, and sensitivity differ.Algorithms reported accuracy of 82.9%/98.27%, specificity of 80.5%/97.60%, and sensitivity of 84%/98.93%, but used datasets with different sample counts.
  • Dataset standardization: Collaborative contributions from governments, firms, and health organizations could provide large, high-quality datasets from clinical and digital sources.Potential sources include hospital X-ray and CT scans, satellite data, personal information, and self-diagnosis reports.
  • Collaborative data development: An Alibaba DAMO Academy collaboration with Chinese hospitals used CT scans from more than 5,000 confirmed cases to build detection systems.The system was reported to achieve 96% accuracy within 20 seconds and had been used by more than 20 hospitals in China.
  • Privacy and security: Privacy and security create a trade-off because effective AI and big-data platforms require personal information that people may not want to share.Relevant data include GPS locations, CT scans, diagnosis reports, travel trajectories, and daily activities.
  • Potential responses: Blockchain, federated learning, and incentive mechanisms are presented as possible ways to address data privacy, decentralized training, and dataset participation.Federated learning keeps most personal data from being transmitted centrally, while incentives aim to increase data quantity and quality.

B. LESSONS AND RECOMMENDATIONS

The survey concludes that AI and big data support applications from outbreak tracking and virus detection to diagnosis and treatment. It recommends improving analytical reliability, combining these technologies with emerging systems, and pursuing broader lessons for pandemic response.

  • Lessons and recommendations: Algorithms should be further optimized to improve the accuracy and reliability of data analytics for COVID-19 diagnosis and treatment.This recommendation is presented as a way to strengthen AI- and big-data-based approaches.
  • Lessons and recommendations: Combining AI and big data with emerging technologies could produce additional solutions for combating COVID-19.Examples include cloud-based vaccine design and 5G-enabled drones, IoT, and localization for sample delivery, goods transport, distancing, and movement monitoring.
  • Lessons and recommendations: AI and big data technologies support COVID-19 applications including outbreak tracking, virus detection, diagnosis, treatment, and drug or vaccine discovery.The survey reviews AI applications in detection, prediction, infodemiology, biomedicine, and pharmacotherapy, alongside related big-data applications.
  • Lessons and recommendations: The survey highlights challenges, lessons, and recommendations for authorities and research communities seeking to improve pandemic responses.Its coverage spans AI and big-data applications, associated challenges, and guidance for future action.
Loading 2107.14040v1…