Dr-SAM: An End-to-End Framework for Vascular Segmentation, Diameter Estimation, and Anomaly Detection on Angiography Images

TL;DR

Dr-SAM addresses the need for automated, end-to-end analysis of peripheral vessels in angiography by integrating a customized positive-point prompting strategy for the Segment Anything Model, skeleton-based diameter estimation via distance transform, and extremum-based anomaly detection. It introduces a new benchmark dataset of pelvic-iliac angiography images with expert annotations and anomaly labels. The approach demonstrates superior segmentation performance over baseline SAM configurations and provides robust stenosis/aneurysm detection with efficient computation, highlighting potential for faster, more accurate clinical decision-making in vascular disease management.

Abstract

Recent advancements in AI have significantly transformed medical imaging, particularly in angiography, by enhancing diagnostic precision and patient care. However existing works are limited in analyzing the aorta and iliac arteries, above all for vascular anomaly detection and characterization. To close this gap, we propose Dr-SAM, a comprehensive multi-stage framework for vessel segmentation, diameter estimation, and anomaly analysis aiming to examine the peripheral vessels through angiography images. For segmentation we introduce a customized positive/negative point selection mechanism applied on top of the Segment Anything Model (SAM), specifically for medical (Angiography) images. Then we propose a morphological approach to determine the vessel diameters followed by our histogram-driven anomaly detection approach. Moreover, we introduce a new benchmark dataset for the comprehensive analysis of peripheral vessel angiography images which we hope can boost the upcoming research in this direction leading to enhanced diagnostic precision and ultimately better health outcomes for individuals facing vascular issues.

Figures (6)

• Figure 1: As an end-to-end framework for angiography image analysis, Dr-SAM first extracts the segments of blood vessels, then detects the centerlines and estimates the diameters of the vessels (shown with circles), then, finally, recognizes the anomaly points indicating stenoses or aneurysms (green points).
• Figure 2: SAM result on X-Ray image without any prompts. Middle - ground truth mask, right - SAM predicted mask.
• Figure 3: The overview of Dr-SAM: For each bounding box provided by the user, first we determine five positive points using our point finder algorithm. This is followed by vessel extraction by SAM SAM conditioned on our positive points. Then for anomaly detection we extract the centerline of the binary mask, obtained from the previous stage, by finding its skeleton, and use that skeleton for estimating vessel diameters, which are later being used to detect anomaly points on the vessels.
• Figure 4: Qualitative comparison of three different approaches for segmentation. From left to right: SAM, naive approach of selecting the positive point as the most probable vascular pixel, our method
• Figure 5: Dr. SAM pipeline results for each step