EDF-Seg: Enhanced Differential Feature Guided Segmentation of Incremental Blood Flow in Coronary Artery

Coronary angiography is the primary method for diagnosing coronary artery disease (CAD). By analyzing the incremental blood flow in coronary artery, we can identify vascular structures and calculate blood flow velocity. Conventional segmentation methods are susceptible to various factors, including noise, artifacts, and vascular motion. These factors often result in unstable segmentation outcomes and consequently compromise the accuracy of blood flow velocity. In this study, we propose a novel coronary angiography incremental blood flow segmentation model, termed EDF-Seg, which is based on enhanced differential feature. The model aims to predict incremental blood flow in coronary artery accurately by analyzing morphological variations in coronary angiography images. To overcome the temporal discontinuity in frame-to-frame features, the differential feature enhancement module proposed in this study first extracts the difference features between two sequential images. Specifically, the module first performs weighted operations to the extracted difference features. Subsequently, it employs an attention mechanism to capture long-range dependencies within the incremental regions. Finally, it integrates multi-scale feature information and generates an inter-frame incremental feature map, thereby highlighting the temporal variations of vascular structures. The method proposed in this study was evaluated on a dataset of right coronary angiography images. The EDF-Seg model achieved an F1 score of 74.3%, representing a 5.5% improvement over change detection methods in the natural image domain. This advancement provides clinicians with an accurate tool for calculating blood flow velocity and holds significant clinical application value.