Gaussian Spatial Priors for Anatomy-Aware Object Detection in Surgical Videos
Researchers have developed a new method called Gaussian Spatial Prior (GSP) to improve object detection in surgical videos, specifically for identifying anatomical structures during inguinal hernia repair. This GSP module injects a parametric bias into the self-attention mechanism of the DAB-DETR model, leveraging the anatomically constrained spatial relationships between structures. The GSP module significantly enhances the detection of smaller, visually ambiguous structures like epigastric vessels, outperforming existing methods like DAB-DETR and YOLOv26. AI
IMPACT Enhances precision in surgical AI applications by improving anatomical structure identification.