Deep Learning Approaches for COVID-19 Detection: A Brief Review of Datasets, Techniques, and Challenges
Résumé
The COVID-19 pandemic has underscored the critical need for rapid and accurate diagnostic tools to curb its spread and impact. Deep learning (DL) techniques have emerged as a promising solution, leveraging advancements in artificial intelligence to analyze medical imaging data, such as chest X-rays and computed tomography (CT) scans, for automated detection and diagnosis of COVID-19. This paper reviews the current state of DL-based approaches for COVID-19 detection, highlighting their potential and limitations. Despite notable successes in experimental settings, several challenges impede the deployment of DL models in clinical environments. These include the scarcity of high-quality and diverse datasets, issues with class imbalance, and inconsistencies in imaging protocols, all of which limit model generalizability. Furthermore, the "black-box" nature of many DL models raises concerns about interpretability and trust, particularly in high-stakes medical decision-making. Additionally, regulatory and clinical validation hurdles pose significant barriers to the widespread adoption of these models. To address these challenges, the paper explores emerging research directions, including synthetic data generation, multi-modal learning, federated learning for privacy-preserving model training, and advancements in Explainable AI (XAI) techniques. The integration of DL models into clinical workflows is also discussed, emphasizing the importance of usability, interoperability, and clinician training to ensure seamless adoption. While significant obstacles remain, the progress made thus far demonstrates the potential of deep learning to revolutionize COVID-19 diagnostics and, more broadly, medical imaging and healthcare. By addressing the outlined challenges and fostering collaboration among researchers, clinicians, and policymakers, AI-driven diagnostic tools can be transformed from experimental innovations into reliable, impactful solutions for global health challenges.
| Origine | Fichiers produits par l'(les) auteur(s) |
|---|---|
| Licence |