Deep learning for heart failure prediction from chest X-ray in AF
Journal
Array
Series/Report No.
Array
Journal Volume
30
Pages
100874
ISSN
2590-0056
Date Issued
2026-07-01
Author(s)
Tsai, Dai-Hua
Faisal, Muhamad
Leu, Jenq-Shiou
Abstract
Heart failure (HF) and atrial fibrillation (AF) often coexist in elderly patients. This study assessed the feasibility of artificial intelligence (AI)-driven chest X-ray (CXR) analysis for HF detection in AF patients. Using convolutional neural networks, 1840 CXRs from AF patients with HF and 2100 without HF were analyzed. Among tested models, Inception-ResNet-v2 demonstrated the highest accuracy (91.2%), with precision (94.9%), recall (87.9%), and F1 score (91.2%). The model achieved an area under the curve of 0.968, with sensitivity and specificity of 0.864 and 0.967, respectively. Given the challenge of diagnosing HF from CXR alone, these findings highlight the potential of AI-based analysis to enhance early detection and intervention in AF patients. By improving diagnostic accuracy and accessibility, AI-driven imaging may support clinical decision-making and optimize patient outcomes.
Publisher
Elsevier BV
Type
journal