Clinically Aligned Geometry Constraints for Robust IVUS Vessel Boundary Segmentation

2026-06-17 · Source: Computer Vision and Pattern Recognition · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Medical Imaging AI · Depth: Expert, quick

Summary

GeoCat is a geometry-consistent neural network designed to improve intravascular ultrasound (IVUS) lumen and external elastic membrane (EEM) segmentation, crucial for accurate coronary plaque burden assessment. Standard segmentation methods often introduce boundary drift and topology errors, leading to inaccurate clinical measurements. GeoCat processes 5-frame IVUS clips using dual Cartesian-polar encoders with cross-domain attention and temporal fusion. It employs a differentiable geometry consistency loss that directly supervises clinically relevant descriptors such as diameters, orientations, and cross-sectional areas. The model was trained on 12,242 annotated frames from 146 patients across two commercial IVUS systems. GeoCat achieved a Dice score of 0.93, reduced 95HD to 0.14 mm, and lowered topology violations to 1.0%. Critically, it significantly improved geometric fidelity, demonstrating diameter errors of 0.13-0.16 mm and angular errors of approximately 8 degrees, supporting reliable plaque burden quantification.

Key takeaway

For Computer Vision Engineers developing robust IVUS segmentation models, GeoCat's approach offers a path to significantly improve clinical accuracy. You should integrate differentiable geometry consistency losses, supervising metrics like diameters and angles, to reduce boundary drift and topology errors. This method, combined with multi-frame processing and dual-domain encoders, can yield more reliable plaque burden quantification, moving beyond traditional overlap scores for critical diagnostic applications.

Key insights

GeoCat improves IVUS segmentation by integrating geometry consistency loss for clinical accuracy.

Principles

• Clinical geometry constraints enhance segmentation robustness.
• Dual Cartesian-polar encoding improves feature representation.
• Temporal fusion in IVUS clips reduces errors.

Method

GeoCat processes 5-frame IVUS clips using dual Cartesian-polar encoders with cross-domain attention and temporal fusion, supervised by a differentiable geometry consistency loss on clinical descriptors.

In practice

• Apply geometry consistency loss for medical image segmentation.
• Use multi-frame input for temporal context.
• Evaluate segmentation with clinical metrics beyond overlap.

Topics

• Intravascular Ultrasound
• Vessel Segmentation
• Geometry Consistency Loss
• Deep Learning for Medical Imaging
• Coronary Plaque Assessment
• Multi-frame Analysis

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

Related on AIssential

• Towards Real-Time Autonomous Navigation: Transformer-Based Catheter Tip Tracking in Fluoroscopy — A multi-threaded deep learning pipeline enables robust, real-time catheter tip tracking in fluoroscopy for autonomous navigation.
• MVSegNet: A Lightweight Boundary-Aware Network for Fetal Lateral Ventricle Segmentation and Atrial Width Estimation in Prenatal Ultrasound — MVSegNet accurately segments fetal lateral ventricles for atrial width estimation despite ultrasound challenges.
• Align then Refine: Text-Guided 3D Prostate Lesion Segmentation — Text-guided, phase-scheduled training improves 3D prostate lesion segmentation by integrating semantic alignment and confidence-aware refinement.
• HadBalance: A Plug-and-Play Unified Global Geometric Prior Framework for Generalizable Biomedical Segmentation — HadBalance unifies geometric priors for biomedical segmentation, adapting to shape heterogeneity via conflict-aware gradient balancing.
• Spatio-Temporal Fusion Model for Standard View Classification of Echocardiographic Videos — The Spatio-Temporal Fusion Model (STFM) enhances echocardiographic view classification by integrating spatial and temporal data with uncertainty-aware learning.
• Improving Pre-trained Adult Glioma Segmentation Models Using only Post-processing Techniques — Adaptive, radiomic-based post-processing significantly enhances pre-trained glioma segmentation model accuracy without additional GPU resources.

Open in AIssential →

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