GloResNet: A lightweight 3D CNN with global topological features for preterm brain injury prediction

2026-06-01

AI Summary

This study addresses the challenge of automatically predicting brain injury in preterm infants from T2-weighted MRI scans, where limited data availability and high inter-subject variability hinder model performance. To tackle this, the authors propose GloResNet, a lightweight 3D CNN based on ResNet-10 architecture, leveraging MedicalNet pretraining weights and introducing a novel global manifold-preserving z-score normalization strategy to retain image topological structure. The approach further enhances robustness under small-sample conditions through mixup augmentation, class weighting, and test-time augmentation. Evaluated via five-fold cross-validation on the dHCP dataset, GloResNet achieves an average accuracy of 75.18% (peaking at 81.82%), with a specificity of 0.81 and sensitivity of 0.76, demonstrating its effectiveness and clinical potential.

Abstract

This study introduces an automated deep learning framework for predicting brain injury (BI) in preterm infants from T2-weighted MRI (dHCP dataset). We propose GloResNet, a lightweight 3D CNN based on ResNet-10, pretrained on MedicalNet to address data scarcity. A global manifold mapping strategy first resamples each 3D volume to 128x128x128 and then applies subject-wise z-score intensity normalization, thereby preserving global topology while standardizing appearance. Training integrates mixup, class weighting, and test-time augmentation for robustness. In 5-fold cross-validation, GloResNet achieved 75.18% average accuracy (peak 81.82%), with specificity 0.81 and sensitivity 0.76. Results demonstrate that a topology-aware lightweight CNN has the capability to effectively predict neonatal BI, offering a non-invasive screening tool. The source code of this paper can be obtained from the GitHub repository: https://github.com/ICL-SUST/GloResNet-Preterm-Brain

Problem

• brain injury prediction
• preterm infants
• T2-weighted MRI
• neonatal brain injury
• automated screening

Innovation

• lightweight 3D CNN
• global topological features
• preterm brain injury prediction
• manifold mapping
• MedicalNet pretraining