Gaussian Spatial Priors for Anatomy-Aware Object Detection in Surgical Videos

2026-06-13 · Source: Computer Vision and Pattern Recognition · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Robotics & Autonomous Systems, Medical Devices & Health Technology · Depth: Expert, quick

Summary

A new Gaussian Spatial Prior (GSP) module significantly enhances anatomy-aware object detection in surgical videos, particularly for challenging, visually ambiguous structures like epigastric vessels during inguinal hernia repair. Standard methods like DAB-DETR and YOLOv26 struggle with these smaller, intermittently visible structures, though they reliably detect prominent ones such as Cooper's Ligament. The GSP module addresses this by encoding anatomically constrained spatial relationships between structures as a compact, parametric bias. This bias is injected into the self-attention mechanism of a DAB-DETR decoder. The prior is pre-computed offline from training annotations using frozen Gaussian parameters and dynamically recomputed at each decoder layer via iteratively refined reference points. Evaluated on an inguinal hernia repair video dataset with 5-fold cross-validation, GSP improved dependent class detection by +33.5% (AP50) over DAB-DETR and +53.9% over YOLOv26, alongside a +6.0% improvement in anchor detection. These gains were statistically significant ($p=0.012$).

Key takeaway

For Computer Vision Engineers developing intraoperative safety systems, integrate Gaussian Spatial Priors (GSP) into your object detection models. This can significantly improve the reliability of detecting small, ambiguous anatomical structures. The GSP approach boosted detection by +33.5% (AP50) over DAB-DETR. This offers a robust method to enhance surgical AI accuracy, especially for critical structures like epigastric vessels. Evaluate GSP for your next-generation surgical video analysis tools to achieve more dependable anatomical mapping.

Key insights

Anatomical spatial priors significantly improve detection of challenging structures in surgical videos.

Principles

• Anatomical structures have constrained spatial relationships.
• Encoding spatial priors can enhance object detection.
• Iterative refinement of reference points improves accuracy.

Method

The GSP module computes spatial priors offline from annotations as frozen Gaussian parameters, then injects this parametric bias into a DAB-DETR decoder's self-attention, recomputing it at each layer using refined reference points.

In practice

• Apply GSP to improve detection of ambiguous structures.
• Use GSP for intraoperative safety frameworks.
• Integrate GSP with existing DAB-DETR pipelines.

Topics

• Object Detection
• Surgical Video Analysis
• Gaussian Spatial Priors
• DAB-DETR
• Inguinal Hernia Repair
• Medical Imaging AI

Best for: AI Scientist, Computer Vision Engineer, Research Scientist

Related on AIssential

• GeoCFNet: Geometry-Aware Confidence Field Network for Robot-Assisted Endoscopic Submucosal Dissection — GeoCFNet enhances robot-assisted ESD guidance via geometry-aware confidence field estimation.
• ViCrop-Det: Spatial Attention Entropy Guided Cropping for Training-Free Small-Object Detection — ViCrop-Det improves small-object detection by adaptively cropping regions using spatial attention entropy, without retraining.
• NG-GS: NeRF-Guided 3D Gaussian Splatting Segmentation — NG-GS enhances 3D Gaussian Splatting segmentation by integrating NeRF-guided continuous feature fields to resolve boundary artifacts.
• Large-Scale High-Quality 3D Gaussian Head Reconstruction from Multi-View Captures — HeadsUp reconstructs high-quality 3D Gaussian heads from multi-view inputs using a UV-parameterized latent representation.
• Spatial-Temporal Decoupled Adapter for Micro-gesture Online Recognition — Decoupling spatial and temporal adaptation with adaptive augmentation significantly improves micro-gesture recognition in imbalanced datasets.
• FS-DVS: A Frequency-Selective Dynamic Visual Sensing Paradigm for Enhancing Information Completeness — FS-DVS uses a learnable pre-trigger spatial filter to mimic RGCs, enhancing event camera data completeness and noise resilience.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Computer Vision and Pattern Recognition.