A Survey on Applications of Artificial Intelligence in Fighting Against COVID-19

Jianguo Chen, Kenli Li, Zhaolei Zhang, Keqin Li, Philip S. Yu

TL;DR

This paper provides a comprehensive survey of how Artificial Intelligence (AI) has been applied to combat the COVID-19 pandemic. It categorizes AI's contributions across disease detection, virology, drug/vaccine development, and epidemic prediction, highlighting the significant role AI has played and its future potential in public health crises.

What Problem Does This Paper Address?

The COVID-19 pandemic presented an unprecedented global health challenge, demanding rapid and innovative solutions. Traditional methods for disease detection, understanding viral mechanisms, developing treatments, and predicting outbreaks were often too slow or insufficient to keep pace with the virus's rapid spread. This paper addresses the need to systematically review and understand how Artificial Intelligence (AI) technologies have been leveraged to accelerate and enhance various aspects of the fight against COVID-19, providing a roadmap for researchers and practitioners.

Methods

The paper surveys a wide range of AI techniques, primarily focusing on Machine Learning (ML) and Deep Learning (DL). Key methods discussed include:

  • Medical Image Analysis: Convolutional Neural Networks (CNNs), U-Net, V-Net, ResNet, and Inception models for CT and X-ray image segmentation and classification.
  • Virology and Pathogenesis: AI for protein structure prediction (e.g., AlphaFold using ResNet), genomic analysis (e.g., ML for sequence similarity and classification), and drug-target interaction prediction.
  • Drug and Vaccine Development: DL models (e.g., MT-DTI, RL-based RNNs) for small-molecule drug discovery, virtual screening, and predicting binding affinities. AI is also used for epitope prediction in vaccine development (e.g., netMHCpan).
  • Epidemic and Transmission Prediction: Time series analysis, SEIR models, LSTM, GRU, and CNNs for forecasting case numbers, predicting outbreak spread, and analyzing transmission dynamics.
  • Social Control: AI for temperature detection, video tracking, and contact tracing.

The survey also notes the use of techniques like transfer learning and data augmentation to address data scarcity.

Datasets

The paper highlights the critical role of various datasets in AI-driven COVID-19 research. These include:

  • Medical Images: Large collections of CT and X-ray images from sources like Kaggle, Mendeley Data, Coronacases, COVIDx, and specific initiatives like the COVID-19 chest x-ray dataset.
  • Biological Data: Resources such as NCBI, GISAID, PDB, UniProt, ZINC, and ChEMBL for genomic sequences, proteins, drug compounds, and miRNA sequences.
  • Cough Sound Datasets: COUGHVID, Coswara, and others, though noted as scarce.
  • Informatics Resources: COVID-19 situation reports, dashboards, and case data from WHO, CDC, Baidu, Google, and JHU CSSE.

The paper emphasizes the challenge of data imbalance, particularly the scarcity of positive COVID-19 samples compared to negative ones.

Key Insights and Ideas

  • AI's Pervasive Role: AI has been applied across virtually every facet of the COVID-19 response, from initial diagnosis to long-term vaccine development and social policy.
  • Medical Imaging as a Major Battlefield: AI, particularly deep learning, has shown significant promise in accelerating and improving the accuracy of COVID-19 detection and diagnosis through CT and X-ray analysis.
  • Accelerating Drug Discovery: AI is proving invaluable in rapidly screening existing drugs and designing novel compounds, significantly shortening the traditional drug development timeline.
  • Data Scarcity and Imbalance: A major hurdle is the lack of large, well-annotated datasets, especially for positive cases, which impacts the robustness and accuracy of AI models.
  • Interdisciplinary Collaboration is Crucial: Effective AI application in healthcare requires deep collaboration between computer scientists, medical professionals, and domain experts.

Key Results

While the paper is a survey and doesn't present new experimental results, it synthesizes findings from numerous studies:

  • Diagnostic Accuracy: AI models for CT image classification have achieved accuracies ranging from the high 70s to over 99%, with sensitivities and specificities often exceeding 90%.
  • Drug Discovery: AI has identified potential drug candidates and accelerated the process of finding inhibitors for key viral proteins like 3CLpro.
  • Epidemic Prediction: AI models have demonstrated the ability to predict outbreak trends and transmission routes, with some systems providing early warnings. For instance, BlueDot predicted the outbreak days before official announcements.

Limitations and Future Work

Limitations:

  • Data Scarcity and Imbalance: The rapid nature of the pandemic led to limited availability of large, balanced datasets, hindering model performance.
  • Data Noise and Rumors: AI struggles to filter misinformation and fake news prevalent on social media, impacting epidemic prediction accuracy.
  • Interdisciplinary Knowledge Gap: A disconnect exists between AI expertise and medical domain knowledge, requiring better integration.
  • Data Privacy: The collection of sensitive personal data for AI applications raises significant privacy and human rights concerns.

Future Work:

The authors suggest several promising directions:

  • Non-contact Disease Detection: AI for automated image acquisition and patient positioning to reduce infection risk.
  • Few-Shot and Transfer Learning: To address data imbalance and scarcity.
  • Remote Video Diagnosis: AI-powered systems for remote consultations and preliminary diagnoses.
  • Patient Prognosis Management: AI for long-term patient monitoring and behavior analysis.
  • Enhanced Biological Research: AI for discovering viral protein structures and gene sequences.
  • Improved Drug and Vaccine Development: AI for simulating interactions and predicting patient responses.
  • Fake News Identification: AI to combat misinformation on social media.
  • Impact Simulation: AI for evaluating social control strategies.
  • Advanced Contact Tracking: AI for building social networks to predict spread.
  • Intelligent Robots: For disinfection, delivery, and patient care.
  • AI-Integrated IoT: For intelligent monitoring in public spaces.

Summary updated

BibTeX
@article{chen2022survey,
  author = {Jianguo Chen and Kenli Li and Zhaolei Zhang and Keqin Li and Philip S. Yu},
  title = {{A Survey on Applications of Artificial Intelligence in Fighting Against COVID-19}},
  journal = {ACM Computing Surveys},
  year = {2022},
  doi = {10.1145/3465398},
}
Figure