Introduction and objective:
The COVID-19 pandemic accelerated development of multimodal artificial intelligence (AI) models that combine chest computed tomography (CT), chest X-ray (CXR) and clinical/laboratory data to support imaging-based diagnosis, triage and prognostication. This review synthesizes reported performance, clinical utility and limitations.

Review methods:
PubMed, Scopus and Google Scholar (Jan 2020–Jun 2025) were searched and included peer-reviewed English studies applying machine learning or deep learning to CT and/or CXR with reported sample sizes and performance metrics (AUC, sensitivity, specificity, F1). Preprints, case reports and studies without sample sizes or performance metrics were excluded.

Brief description of the state of knowledge:
Meta-analyses report high discriminative performance for severity prediction (pooled AUC ≈ 0.89) alongside a high prevalence of study bias. Selected studies reported top accuracies up to ≈98% and multimodal F1 scores up to 0.89. Recurring limitations were dataset heterogeneity, single-centre training and scarce external validation. Some reports found automated pipelines substantially faster than manual reads (e.g. ~2.7 s vs ~6.5 min), although workflow times vary by setting.

Summary:
Multimodal integration of CT, CXR and clinical data with AI is promising for rapid, reproducible assessment of COVID-19 severity. Clinical translation requires standardized acquisition and reporting, rigorous multicentre external validation, transparent methods, and formal evaluation of clinical impact and fairness.