Facilitating Timely Decision-Making in Healthcare: An Object Detection Approach for Automated Coronary Artery Stenosis Detection

Hadis  Keshavarz; Hossein  Sadr; Mojdeh  Nazari; Arsalan  Salari

doi:10.18502/fbt.v13i2.21947

Hadis Keshavarz Cardiovascular Diseases Research Center, Department of Cardiology, Heshmat Hospital, School of Medicine, Guilan University of Medical Sciences, Rasht, Iran
Hossein Sadr Guilan Road Trauma Research Center, Trauma Institute, Guilan University of Medical Sciences, Rasht, Iran
Mojdeh Nazari Cardiovascular Diseases Research Center, Department of Cardiology, Heshmat Hospital, School of Medicine, Guilan University of Medical Sciences, Rasht, Iran
Arsalan Salari Cardiovascular Diseases Research Center, Department of Cardiology, Heshmat Hospital, School of Medicine, Guilan University of Medical Sciences, Rasht, Iran

DOI: https://doi.org/10.18502/fbt.v13i2.21947

Keywords: Object Detection; Deep Learning; Medical Image; Coronary Angiography; Digital Medicine.

Abstract

Purpose: Coronary Artery Disease (CAD), characterized by coronary artery stenosis—the narrowing of arteries supplying blood to the heart—is a leading global cause of morbidity and mortality. Timely detection and management of stenosis are crucial to preventing severe outcomes such as myocardial infarction and heart failure. Despite advancements in medical imaging, current diagnostic methods rely heavily on the manual interpretation of coronary angiograms, which is time-consuming, subjective, and prone to variability. To address these limitations, this study proposes an automated object detection-based framework for identifying coronary artery stenosis in medical imaging.

Materials and Methods: The study employs two state-of-the-art deep learning models, RetinaNet and EfficientDet D3, to detect stenotic regions in X-ray angiography images. A dataset of 8,325 annotated images from 100 patients with single-vessel CAD, sourced from the Research Institute for Complex Issues of Cardiovascular Diseases in Kemerovo, Russia, was used for training and evaluation. To enhance model performance, a comprehensive preprocessing pipeline was applied, including image resizing, data augmentation, and intensity normalization. These steps ensured robustness and generalizability across diverse imaging conditions.

Results: Both models demonstrated high accuracy in stenosis detection. RetinaNet achieved a mean Average Precision (mAP) of 93.2%, while EfficientDet D3 outperformed with an mAP of 96.6%. These results highlight the models' ability to accurately identify stenosis, even in noisy and variable angiographic images. The superior performance of EfficientDet D3 underscores its potential for clinical integration, offering precise and reliable stenosis localization.

Conclusion: This study presents a robust and efficient deep learning framework for the automated detection of coronary artery stenosis. By reducing reliance on manual interpretation and enhancing diagnostic accuracy, the proposed approach supports timely and informed clinical decision-making. This innovation has the potential to streamline diagnostic workflows, improve patient outcomes, and advance the application of artificial intelligence in cardiovascular healthcare.