GLeVE: Graph-Guided Lesion Grounding with Proposal Verification in 3D CT

2026-05-21 · Source: Computer Vision and Pattern Recognition · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Health & Medical Research · Depth: Expert, quick

Summary

GLeVE is a novel graph-guided lesion grounding framework designed to address the semantic-spatial gap in linking radiology report descriptions to 3D CT volumes. It improves upon existing phrase-level alignment and dense pixel supervision methods by focusing on lesion-wise correspondence and localization accuracy. GLeVE employs relation-aware graph reasoning to encode organ attribution, attributes, and inter-lesion relations, generating discriminative lesion-wise queries. The framework integrates anatomy-aware proposal generation with region-level verification to ensure one-to-one text-lesion alignment, alongside hierarchical octree refinement for precise boundary delineation. Experiments on AbdomenAtlas 3.0 demonstrate GLeVE's consistent gains in both segmentation accuracy and lesion-level localization compared to classical multimodal foundation models and report-supervised baselines.

Key takeaway

For Research Scientists developing medical image analysis systems, GLeVE offers a robust approach to improve lesion grounding in 3D CT volumes. You should consider integrating graph-guided reasoning and anatomical prior verification into your models to overcome semantic-spatial gaps and achieve more accurate lesion-wise correspondence and segmentation. This framework's success on AbdomenAtlas 3.0 suggests a path for enhancing verifiable clinical interpretation.

Key insights

GLeVE uses graph reasoning and anatomical verification to precisely ground radiology report lesions in 3D CT scans.

Principles

• Relation-aware graph reasoning improves lesion queries.
• Anatomy-aware verification ensures one-to-one alignment.
• Hierarchical octree refinement enhances boundary delineation.

Method

GLeVE encodes organ attribution and inter-lesion relations via graph reasoning, generates anatomy-aware proposals with region-level verification for alignment, and refines boundaries using hierarchical octree processing.

In practice

• Apply graph reasoning for complex medical text-image alignment.
• Integrate anatomical priors for robust lesion verification.
• Utilize octree refinement for precise 3D segmentation.

Topics

• Lesion Grounding
• 3D CT Imaging
• Radiology Reports
• Graph Reasoning
• Medical Image Segmentation
• Octree Networks

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

Related on AIssential

• Scalable Training of Spatially Grounded 2D Vision-Language Models for Radiology — Spatially grounded radiology VLMs can be trained at scale without manual annotations, maintaining language quality.
• Geometry-Aware Superpixel Graph Transformer with Metadata for Skin Lesion Classification — Explicitly modeling skin lesions as geometry-aware superpixel graphs with integrated metadata improves classification accuracy.
• CA-GCL: Cross-Anatomy Global-Local Contrastive Learning for Robust 3D Medical Image Understanding — CA-GCL uses global-local contrastive learning and clinical-aware text augmentation to prevent representation collapse in 3D medical FVLP.
• 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.
• Pseudo-Text-Conditioned 3D Grounding DINO for Organ Localization in Abdominal CT — A lightweight 3D detector adapts Grounding DINO using frozen pseudo-text tokens for organ localization in abdominal CT.
• Enhancing Brain MRI Anomaly Detection and Reasoning with ROI Rethink and Synthetic Data — Explicit hypothesis-verification grounding with ROI marking enhances trustworthy brain MRI anomaly detection and reduces hallucination.

Open in AIssential →

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