Improving Generalization of Deep Learning for Brain Metastases Segmentation Across Institutions
Summary: arXiv:2604.00397v1 Announce Type: cross
Abstract:
Background: Deep learning has demonstrated significant potential for automated brain metastases (BM) segmentation; however, models trained at a singular institution often exhibit suboptimal performance at various sites due to disparities in scanner hardware, imaging protocols, and patient demographics. The goal of this work is to create a domain adaptation framework that will allow for BM segmentation to be used across multiple institutions.
Methods
We propose a VAE-MMD preprocessing pipeline that combines variational autoencoders (VAE) with maximum mean discrepancy (MMD) loss, incorporating skip connections and self-attention mechanisms alongside nnU-Net segmentation. The method was tested on 740 patients from four public databases: Stanford, UCSF, UCLM, and PKG, evaluated by domain classifier’s accuracy, sensitivity, precision, F1/F2 scores, surface Dice (sDice), and 95th percentile Hausdorff distance (HD95).
Results
The VAE-MMD approach achieved notable improvements in BM segmentation metrics. The results are summarized as follows:
- Domain classifier accuracy reduced from 0.91 to 0.50, indicating successful feature alignment across institutions.
- Reconstructed volumes attained a Peak Signal-to-Noise Ratio (PSNR) greater than 36 dB, ensuring anatomical accuracy.
- The mean F1 score increased by 11.1% (from 0.700 to 0.778).
- The mean surface Dice (sDice) improved by 7.93% (from 0.7121 to 0.7686).
- The mean Hausdorff distance (HD95) decreased by 65.5% (from 11.33 mm to 3.91 mm) across all four centers compared to the baseline nnU-Net.
Conclusions
The VAE-MMD framework effectively diminishes cross-institutional data heterogeneity and enhances BM segmentation generalization across volumetric, detection, and boundary-level metrics without necessitating target-domain labels. This advancement overcomes a significant obstacle to the clinical implementation of AI-assisted segmentation, paving the way for improved patient outcomes and streamlined practices in medical imaging.
The implications of this study are profound, suggesting that with proper adaptation techniques, deep learning models can be more universally applied, thus improving the scalability and reliability of automated segmentation tools in clinical settings.