Source-linked AI summary
Artificial Intelligence in the Battle against Coronavirus (COVID-19): A Survey and Future Research Directions
Thanh Thi Nguyen, Quoc Viet Hung Nguyen, Dung Tien Nguyen, Samuel Yang, Peter W. Eklund, Thien Huynh-The, Thanh Tam Nguyen, Quoc-Viet Pham, Imran Razzak, Edbert B. Hsu
TL;DR
COVID-19 created a global crisis requiring methods for diagnosis, modelling, information analysis, and biomedical research. This paper surveys AI applications across these areas, compiles relevant data sources, and discusses future research directions. It concludes that AI has contributed broadly to COVID-19 responses, but scarce data, explainability, validation, and bias remain important constraints.
Problem
The paper addresses the need to understand how AI can support diverse COVID-19 response and control problems across medical, epidemiological, informational, and biomedical domains.
Method
The authors conduct a comprehensive survey organized by AI application domains, covering medical imaging, pandemic modelling, IoT, text mining and NLP, and computational biology and medicine.
Results
The survey identifies broad AI applications in COVID-19 diagnosis, transmission modelling and forecasting, text mining and NLP, IoT, and computational biology and medicine.
Takeaways & Limitations
The paper provides an overview of current AI applications and identifies research directions for addressing COVID-19-related problems as more relevant data become available.
Takeaways & Limitations
Current AI applications face scarce or poor-quality data, limited explainability, insufficient external validation, reproducibility problems, and bias that constrain clinical translation.
Abstract
from arXiv · showhide
Artificial intelligence (AI) has been applied widely in our daily lives in a variety of ways with numerous success stories. AI has also contributed to dealing with the coronavirus disease (COVID-19) pandemic, which has been happening around the globe. This paper presents a survey of AI methods being used in various applications in the fight against the COVID-19 outbreak and outlines the crucial role of AI research in this unprecedented battle. We touch on areas where AI plays as an essential component, from medical image processing, data analytics, text mining and natural language processing, the Internet of Things, to computational biology and medicine. A summary of COVID-19 related data sources that are available for research purposes is also presented. Research directions on exploring the potential of AI and enhancing its capability and power in the pandemic battle are thoroughly discussed. We identify 13 groups of problems related to the COVID-19 pandemic and highlight promising AI methods and tools that can be used to address these problems. It is envisaged that this study will provide AI researchers and the wider community with an overview of the current status of AI applications, and motivate researchers to harness AI's potential in the fight against COVID-19.
1 Introduction
COVID-19 spread globally with severe human consequences, prompting technological responses including robotic and drone assistance. The paper situates AI as a broad toolkit for pandemic applications, surveying methods and organizing prior work by application domains.
- Pandemic context: COVID-19 spread to almost every country, with over 6,043,000 deaths among nearly 456,798,000 confirmed cases by mid-March 2022.Governments proposed intervention policies, while robots and drones supported hospital services and disinfection efforts.
- AI applications: Computational techniques can process medical images such as X-rays and CT scans to support early detection of infectious patients.
- Survey scope: This survey broadens prior technique- and aspect-based reviews by organizing AI applications according to domains supporting COVID-19 response and control.The paper contrasts categorization by computational-intelligence techniques or prediction, symptom recognition, and development with its application-domain organization.
- AI capabilities: Common AI methods include CNN architectures for computer vision and autoencoders and recurrent neural networks for prominent NLP tools.Examples of CNN architectures include LeNet, AlexNet, GoogLeNet, VGG Net, and ResNet.
- AI capabilities: Machine learning, especially deep learning, has advanced computer vision, NLP, speech recognition, and computer games while handling large, unstructured data types.Examples include text, images, video, and audio.
2 AI against COVID-19: A Survey
The survey organizes AI applications against COVID-19 across medical imaging, pandemic modelling, IoT, text mining and NLP, and computational biology and medicine. Reported studies illustrate AI for detection, forecasting, risk assessment, molecular design, and genomic analysis, while highlighting the need for comparable clinical benchmarks.
- The survey categorizes COVID-19 AI research into five application domains spanning medical imaging, pandemic modelling, IoT, text mining and NLP, and computational biology.These categories organize the surveyed literature by prominent application domain.
- Medical Image Processing with Deep Learning: COVNet uses volumetric chest CT images and a convolutional ResNet-50 model to classify non-pneumonia, community acquired pneumonia, and COVID-19.Its AUC is 0.96 on CT images collected from six hospitals in China.
- Medical Image Processing with Deep Learning: A location-attention mechanism concatenated with three-dimensional CNN ResNet-18 achieves approximately 86% overall accuracy on COVID-19, influenza-A pneumonia, and healthy cases.The approach exploits distinct CT manifestations to support early screening.
- AI-based Data Science Methods for COVID-19 Modelling: Pandemic modelling supports disease-status assessment, intervention evaluation, response planning, infection forecasting, and medical-resource allocation.Forecasting can inform decisions about controls and resources including intensive care beds, staff, ventilators, therapeutics, and vaccines.
- AI-based Data Science Methods for COVID-19 Modelling: Surveyed modelling and sensing systems combine deep learning, heterogeneous real-time data, epidemic models, NLP, smartphones, and agent-based simulation for warnings, risk assessment, forecasting, and tracing.Examples include a stacked autoencoder for confirmed-case warnings, α-Satellite for geographical infectious-risk assessment, and hybrid SI–NLP–deep-learning forecasting.
- AI in Computational Biology and Medicine: Computational biology applications use generative models for drug-like molecule design and machine learning for viral taxonomy and host-origin analysis.Analyses reported taxonomy consistent with Sarbecovirus and quantitative evidence supporting bat and pangolin host hypotheses.
3 COVID-19 Data Sources and Potential Modelling Approaches
The paper catalogs COVID-19 data sources across numerical, radiological, textual, social-media, and biological-sequence formats, linking each type to suitable AI modelling approaches. It emphasizes hybrid methods and standardized, accessible data resources for advancing research.
- Data sources: COVID-19 research data span infection-case numbers, radiology images, Twitter, text, natural language, and biological sequences.These sources are downloadable and collected to support AI research from multiple perspectives.
- Modelling approaches: Different data characteristics require different AI methods, while hybrid models can combine complementary strengths and reduce individual weaknesses.The paper describes hybrid approaches as promising for handling varied COVID-19 data.
- Modelling approaches: CNN-based deep-learning models are commonly suited to high-dimensional chest X-ray and CT radiology images.Examples include LeNet, AlexNet, GoogLeNet, VGG Net, and ResNet.
- Modelling approaches: Natural-language data require text-mining and advanced NLP or NLG tools for classification, summarization, translation, recognition, and question answering.The listed tools include NLTK, ELMo, ULMFiT, and Transformer-based models.
- Modelling approaches: Traditional machine learning, deep learning, or combinations of both can analyze viral genomic and proteomic sequences, depending on the task and data pipeline.Hierarchical clustering and DBSCAN are examples for investigating virus origin from genomic sequences.
4 Recommendations and Future Research Directions
The paper identifies methodological, data, and deployment challenges that constrain COVID-19 AI applications, and proposes research directions spanning clinical validation, standardized datasets, explainability, and diverse application domains. It organizes 13 problem groups by data types, application domains, and suitable AI methods.
- Clinical validation and interpretability: Radiology-based ML studies commonly lack reproducibility, sufficient external validation, and freedom from data or development bias, limiting clinical translation.The paper states that addressing these problems together is necessary for adoption into future clinical practice.
- Clinical validation and interpretability: Explainable AI is needed to reveal which imaging features drive predictions and provide clinicians with interpretable evidence for distinguishing COVID-19 from other pneumonia.The paper connects improved interpretability with more effective examination of future coronavirus CT and X-ray images.
- Data infrastructure and benchmarking: Creating, hosting, and benchmarking datasets under standardized protocols is essential because current COVID-19 datasets use varied formats and standards.The proposed repositories would allow researchers worldwide to contribute to and freely use datasets for research.
- Data infrastructure and benchmarking: A benchmark framework should validate and compare deep learning methods across clinical imaging datasets using consistent computing hardware and evaluation conditions.The paper presents benchmarking as a response to the difficulty of translating heterogeneous studies into real-world clinical practice.
- Application directions: AI research directions cover surveillance, public-policy analysis, chatbots, misinformation, text mining, EHR prediction, computational biology, and drug discovery.Examples include tracking spread, analyzing intervention responses and mental health, predicting clinical outcomes, characterizing viral signatures, and supporting vaccine or drug development.
- Application directions: Table 3 identifies 13 COVID-19 problem groups and matches each with needed data, potential AI methods, challenges, and related work.The paper recommends traditional ML or LSTM for clinical and time-series data, CNNs for images, sequence-alignment methods for viral genomes and proteins, and NLG/NLP tools for text.
5 Conclusions
The survey finds AI applied across diagnosis, pandemic modelling, text mining, NLP, IoT, and computational biology, but its contributions remain relatively limited. Scarce COVID-19 data constrain current progress, while data growth and responsible system development are identified as important future needs.
- AI applications span medical diagnosis, transmission modelling and forecasting, public-awareness analysis, and biological data analysis for drug discovery.
- The survey reports that AI contributions to the COVID-19 response remain relatively limited, partly because COVID-19 data are scarce.
- More available medical images and biological sequences are expected to increase the number of COVID-19-related AI studies.
- AI systems face challenges involving poor training and validation data, explainability, and the resulting trust deficit.
- Effective deployment requires data pipelines, model selection, retraining, continuous monitoring, validation, and adherence to ethics principles.