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

1

An automated deep learning framework was developed for predicting hematoma expansion in spontaneous intracerebral hemorrhage using non-contrast CT scans.
2

The framework includes automated segmentation, synthetic data augmentation, and Vision Transformer-based classification for improved prediction accuracy.
3

The ViT-1:2 classifier achieved an AUC of 0.815 on the training set and demonstrated strong external validation AUCs of 0.793 and 0.781.
4

Hematoma expansion occurs in 20-30% of sICH patients, significantly impacting mortality and long-term disability, necessitating rapid intervention.
5

The U-Mamba model was identified as the optimal architecture for segmentation, achieving high accuracy and robustness in challenging clinical scenarios.

Npj Digital Medicine

by Qiang Yu, Xin Fan, Jinwei Li, Qianyu Hao, Youquan Ning, Shichao Long, Wenhao Jiang, Fajin Lv, Xianlei Yan, Quan Liu, Xiaoquan Xu, Zongqian Wu, Juan Peng, Min Wu
December 7, 2025

1