Anatomically-guided Masked Autoencoder with Domain-Adaptive Prompting (AMAP) for multimodal cerebral aneurysm detection and segmentation - Summary - MDSpire

Anatomically-guided Masked Autoencoder with Domain-Adaptive Prompting (AMAP) for multimodal cerebral aneurysm detection and segmentation

  • By

  • Mingxuan Huang

  • Tiantian Liu

  • Jiayin Zhang

  • Xiaoming Su

  • Hanlin Chen

  • Miao Li

  • Jinghan Guo

  • Kaiyang Zu

  • Xiaofeng Chen

  • Yanguo Su

  • Hengri Cong

  • Long Yan

  • Tianyi Yan

  • Yiming Deng

  • December 8, 2025

  • 0 min

Share

Objective:

To develop a robust framework for the detection and segmentation of cerebral aneurysms, particularly small ones, which are critical for risk assessment and treatment planning.

Key Findings:
  • AMAP outperforms CNN, Transformer, and foundation-based baselines on multiple datasets, achieving 3−5% higher Dice scores and reducing false positives per case by about 20%.
  • Demonstrates accurate boundary preservation and consistent detection of small aneurysms, which are often overlooked by existing methods.
Interpretation:

AMAP represents a significant advancement in AI for cerebral aneurysm detection, with practical implications for improving clinical outcomes in identifying small aneurysms and enhancing domain generalization.

Limitations:
  • Performance may vary with different imaging modalities not included in the training datasets; future work should explore additional modalities.
  • Further validation is needed in clinical settings to confirm generalizability and effectiveness across diverse patient populations.
Conclusion:

AMAP is a promising step towards reliable AI-assisted aneurysm screening, enhancing both detection accuracy and interpretability in clinical practice.

Original Source(s)

Anatomically-guided Masked Autoencoder with Domain-Adaptive Prompting (AMAP) for multimodal cerebral aneurysm detection and segmentation

  • by Mingxuan Huang, Tiantian Liu, Jiayin Zhang, Xiaoming Su, Hanlin Chen, Miao Li, Jinghan Guo, Kaiyang Zu, Xiaofeng Chen, Yanguo Su, Hengri Cong, Long Yan, Tianyi Yan, Yiming Deng

  • December 8, 2025

Related Content

A generalizable 3D framework and model for self-supervised learning in medical imaging

An end-to-end deep learning pipeline for hematoma expansion prediction in spontaneous intracerebral hemorrhage based on non-contrast computed tomography

Deep learning model for assessing survival benefits in hepatocellular carcinoma patients undergoing intra-arterial therapies based on proliferative subtype