ARTIFICIAL INTELLIGENCE IN PANDEMIC RESPONSE: DIAGNOSTIC STRATEGIES FOR COVID-19

Abstract

The COVID-19 pandemic exposed critical gaps in global diagnostic capacity, surveillance, and rapid clinical decision support. Artificial intelligence (AI) offered a range of diagnostic strategies—ranging from imaging-based models for chest X-ray and CT interpretation to multimodal systems that fuse clinical, laboratory, and digital phenotyping data—for accelerating detection, triage, and resource allocation. This paper synthesizes the state of AI diagnostic methods applied to COVID-19, proposes a unified evaluation framework, and demonstrates a proof-of-concept using a synthetic dataset that simulates common realworld conditions: noisy labels, distribution shift, and limited ground truth. We compare three diagnostic approaches (imaging-only deep convolutional neural networks, clinical-data gradient boosting models, and a multimodal ensemble) on sensitivity, specificity, calibration, and robustness to domain shift. Results on the synthetic benchmark show that multimodal ensemble models achieve the best balance of sensitivity and specificity while maintaining better calibration and robustness across simulated hospitals. We discuss operational challenges—data heterogeneity, bias, regulatory pathways, and clinician adoption—and provide recommendations for integrating AI diagnostics into pandemic response workflows.