Attention-Enhanced Ensemble Learning for Automated Stenosis Detection in X-Ray Coronary Angiography Videos Publisher

Summary: Can AI spot coronary narrowing on angiograms? An attention-based ensemble reached about 78% video-level F1 in a 413-patient dataset, but needs external validation. #Cardiology #MedicalAI

Hosseini M S ; Naghsh Nilchi A R ; Safayani M ; Sadeghi M
Source: 16th International Conference on Information and Knowledge Technology, IKT 2025 Published:2026

Abstract

Coronary artery disease is a major global health concern that requires timely and accurate diagnosis to prevent life-threatening outcomes. While X-ray coronary angiography (XCA) is the clinical gold standard for detecting stenosis, its interpretation remains subjective and variable across observers. This paper presents a deep learning-based framework for automated stenosis detection in XCA videos, focusing on the right coronary artery. From each video, five key frames are extracted using uniform sampling and refined through a segmentation-guided selection method. We employ a stacked ensemble model that combines VGG16 and VGG19 architectures, each enhanced with a Convolutional Block Attention Module to emphasize salient features. Frame-level predictions are aggregated via majority voting to achieve video-level classification. Experimental results on 413 patient cases using five-fold stratified cross-validation show that our model outperforms individual CNN baselines, achieving F1-scores of 77.19% and 77.61% at the frame and video levels, respectively. The results highlight the effectiveness of attention-guided ensemble learning for robust stenosis detection. © 2025 IEEE.