Intracranial Aneurysm Classification and Segmentation via Tri-Axial ROI and Multi-Task Learning

2026-06-25 · Source: Computer Vision and Pattern Recognition · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Computer Vision & Pattern Recognition, Medical Imaging Analysis · Depth: Expert, quick

Summary

A new multi-task framework addresses the limitations of existing automated methods for intracranial aneurysm detection, which typically offer only binary detection. This framework simultaneously performs multi-label classification, multi-class aneurysm segmentation, and multi-class vessel segmentation across 13 anatomical locations and four imaging modalities: CTA, MRA, T2, and T1-post. The two-stage approach integrates a fast 2D tri-axial Region of Interest (ROI) extraction method with a 3D multi-task nnU-Net backbone. Key architectural features include a dual-decoder design to mitigate extreme volume imbalance between aneurysm and vessel classes, alongside cross-attention pooling and modality-specific auxiliary heads for enhanced feature learning across heterogeneous inputs. A two-fold ensemble of this system secured 2nd place in the RSNA 2025 Intracranial Aneurysm Detection challenge. Code, model weights, and a 3D Slicer plugin are publicly available.

Key takeaway

For Computer Vision Engineers developing automated medical image analysis systems for intracranial aneurysms, this multi-task framework offers a robust blueprint. You should consider adopting a two-stage approach with 2D ROI extraction and a 3D multi-task nnU-Net, especially when dealing with diverse imaging modalities and severe class imbalance. Implementing a dual-decoder and cross-attention pooling can significantly improve the accuracy of both classification and fine-grained multi-class segmentation, moving beyond simple binary detection.

Key insights

A multi-task framework enhances intracranial aneurysm analysis by integrating classification and multi-class segmentation across diverse modalities.

Principles

• Multi-task learning improves feature sharing for related medical image analysis tasks.
• Dual-decoder architectures can mitigate extreme class volume imbalance in segmentation.
• Combining 2D ROI extraction with 3D processing optimizes volumetric analysis.

Method

A two-stage process involving fast 2D tri-axial ROI extraction, followed by a 3D multi-task nnU-Net backbone with a dual-decoder, cross-attention pooling, and modality-specific auxiliary heads for classification and segmentation.

In practice

• Implement a dual-decoder for imbalanced multi-class segmentation tasks.
• Utilize cross-attention pooling for integrating heterogeneous imaging modalities.
• Leverage 2D ROI extraction to refine 3D volumetric analysis efficiency.

Topics

• Intracranial Aneurysm
• Multi-Task Learning
• Medical Image Segmentation
• nnU-Net
• Region of Interest
• RSNA Challenge

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

Related on AIssential

• Multi-Class Brain Tumor Classification Using Advanced Deep Learning Models: A Comparative Study — EfficientNetB0 significantly improves multi-class brain tumor classification from MRI images, especially for subtle meningiomas, outperforming other CNNs.
• 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.
• A deep learning framework for glomeruli segmentation with boundary attention — A U-Net-based model with a boundary attention decoder improves glomeruli segmentation in kidney tissue.
• 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.
• Physics constrained graph neural network for real time prediction of intracranial aneurysm hemodynamics — A physics-constrained GNN predicts intracranial aneurysm hemodynamics in real-time, generalizing across patient geometries.
• Leptomeningeal Collateral Detection on DSA via Vessel-Graph Neural Networks — A hybrid graph-pixel neural network enables objective, individualized detection of leptomeningeal collaterals on high-resolution DSA images.

Open in AIssential →

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