ASPSeg: Adaptive Space-Pruning Segmentation for Coronary Angiograms Sequences (Preprint)
Year: 2025
Type: preprint
Abstract: BACKGROUND Image segmentation is essential in analyzing coronary angiography (CAG) X-ray images. However, recent deep learning-based segmentation models are often too complex to be efficiently executed on edge devices. OBJECTIVE This study proposes a novel method called ASPSeg (Adaptive Space-Pruning Segmentation) to ac-celerate the segmentation of CAG sequences. METHODS ASPSeg is a top-down adaptive algorithm that dynamically splits images into patches based on video content, ensuring that computational resources are focused on regions of interest. The performance and accuracy of ASPSeg are enhanced by utilizing spatial-temporal information from video sequences. The method's effectiveness is evaluated on edge nodes. RESULTS The experimental results show that ASPSeg can reduce a CPU’s computational load by nearly 50% and 60%, respectively, with negligible accuracy degradation. CONCLUSIONS ASPSeg is a complementary technique to other model reduction methods, such as pruning and quan-tization, providing an orthogonal approach that improves the efficiency of model inference.
Authors: Kai-Wen Lin, Chih-Kuo Lee, Che-Rung Lee
Language: English
Subfield: Computer Vision and Pattern Recognition
Field: Computer Science
Domain: Physical Sciences
Open Access status: gold
OpenAlex ID: W6963534070