Shape Over Intensity: Directional Topological Encoding for False Positive Reduction in Intracranial Aneurysm Detection

2026-07-08 · Source: cs.CV updates on arXiv.org · Field: Science & Research — Health & Medical Research, Mathematics & Computational Sciences · Depth: Expert, extended

Summary

The Smooth Euler Characteristic Transform (SECT) offers a topology-aware framework to reduce false positives in automated intracranial aneurysm (IA) detection from CT angiography (CTA). Traditional convolutional neural networks struggle with small lesions (<3 mm), often confusing saccular aneurysms with healthy vascular bifurcations due to reliance on local pixel intensities, leading to sensitivities below 60%. SECT, a directional mathematical representation, encodes global 3D vascular geometry independently of intensity. Evaluated on a stratified RSNA 2025 dataset, SECT achieved an AUC of 0.943, significantly surpassing direction-agnostic persistence methods (AUC ~0.68). Crucially, it maintained a 0.943 AUC and 78.5% sensitivity for sub-3 mm aneurysms at 95% specificity. The method also demonstrated high robustness, achieving a 0.927 mean AUC across four distinct scanner manufacturers under leave-one-scanner-out validation. This establishes SECT as a robust, scanner-agnostic filter for deep-learning diagnostic pipelines.

Key takeaway

For Machine Learning Engineers developing automated intracranial aneurysm detection systems, you should integrate topology-aware filters like SECT to drastically reduce false positives. This approach, which prioritizes directional shape over local intensity, significantly improves detection for critical sub-3 mm lesions and ensures robust performance across diverse scanner manufacturers. While current inference times (~11 s/patch) require optimization, incorporating SECT as a downstream filter can enhance clinical viability and trustworthiness of your diagnostic pipelines.

Key insights

Global shape representations, not local intensities, reliably resolve geometric ambiguities in medical image analysis.

Principles

• Local intensity-based models are brittle for structures with similar density profiles.
• Explicit global shape representations resolve geometric ambiguity robustly.
• Topological invariants improve generalizability across varied imaging conditions.

Method

Implement a plug-and-play false-positive reduction filter downstream of CNNs, extracting directional topological features using the Smooth Euler Characteristic Transform (SECT).

In practice

• Integrate SECT into existing deep learning pipelines as a post-processing filter.
• Prioritize global shape analysis for small, geometrically ambiguous lesions.
• Employ directional topological features for cross-scanner robustness.

Topics

• Intracranial Aneurysm Detection
• Topological Data Analysis
• Smooth Euler Characteristic Transform
• False Positive Reduction
• CT Angiography
• Medical Imaging AI
• Scanner Robustness

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

Related on AIssential

• Intracranial Aneurysm Classification and Segmentation via Tri-Axial ROI and Multi-Task Learning — A multi-task framework enhances intracranial aneurysm analysis by integrating classification and multi-class segmentation across diverse modalities.
• Clinically Aligned Geometry Constraints for Robust IVUS Vessel Boundary Segmentation — GeoCat improves IVUS segmentation by integrating geometry consistency loss for clinical 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.
• Landmark-free Assessment of Lower-limb Alignment with Implicit Neural Shape Functions from Knee Radiographs — Implicit Neural Shape Functions enable flexible, landmark-free lower-limb alignment assessment from radiographs, matching leading landmark-based method performance.
• A Dual Edge Spatial Jacobian Image Graph for Interpretable Diabetic Retinopathy Grading — Interpretable DR grading benefits from graph representations linking spatial lesion-vessel geometry and embedding-biomarker sensitivity.
• GloResNet: A lightweight 3D CNN with global topological features for preterm brain injury prediction — Lightweight 3D CNN with global topological features predicts preterm brain injury effectively.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by cs.CV updates on arXiv.org.