# Diagnostic Performance of Artificial Intelligence in Detecting COVID-19 Pneumonia on Chest Imaging

## TL;DR

This paper reviews how AI performs in detecting COVID-19 pneumonia using chest imaging, finding it promising but with limitations in generalizability.

## Contribution

A systematic review of AI diagnostic performance for COVID-19 pneumonia detection using chest imaging, highlighting generalizability challenges.

## Key findings

- CXR-based AI systems showed sensitivities from 80% to 98% and specificities from 82% to 96%.
- CT-based AI models achieved accuracies between 90% and 96%.
- External validation revealed reduced accuracy, indicating generalizability issues.

## Abstract

The COVID-19 pandemic highlighted the need for rapid, accurate, and accessible diagnostic tools. Chest imaging modalities, including chest radiography (CXR) and computed tomography (CT), provided valuable diagnostic information and prompted the development of artificial intelligence (AI) systems to support image interpretation and improve workflow efficiency. This literature review synthesizes current evidence on the diagnostic performance, limitations, and clinical implications of AI models in COVID-19 pneumonia detection through CXR and CT evaluation. A PubMed search was conducted through October 2025 to identify studies evaluating AI systems for the detection of COVID-19 pneumonia using CXR and CT. Studies reporting diagnostic performance metrics, including sensitivity, specificity, accuracy, or area under the curve (AUC), were included. Study quality and risk of bias were assessed using the Quality Assessment of Diagnostic Accuracy Studies-2 (QUADAS-2) tool. Eleven studies met the inclusion criteria. CXR-based AI systems demonstrated sensitivities from 80% to 98% and specificities from 82% to 96%, often comparable to radiologist performance. CT-based AI models achieved accuracies between 90% and 96%. AI models demonstrated strong internal diagnostic performance on CXR and CT but showed reduced accuracy with external validation, underscoring limitations related to generalizability and retrospective study designs. AI models demonstrate promising diagnostic performance for detecting COVID-19 pneumonia on chest imaging and may enhance radiologist efficiency. However, challenges related to generalizability, model adaptability, and clinician trust remain. Future research should prioritize external validation and transparent reporting to ensure the safe and effective integration of AI into clinical practice.