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

2026-06-01 · Source: Takara TLDR - Daily AI Papers · Field: Science & Research — Health & Medical Research, Mathematics & Computational Sciences · Depth: Expert, medium

Summary

GloResNet is a novel lightweight 3D Convolutional Neural Network. It automates brain injury (BI) prediction in preterm infants using T2-weighted MRI scans from the dHCP dataset. Built on a ResNet-10 architecture, GloResNet is pretrained on MedicalNet to address data scarcity. The framework uses a global manifold mapping strategy. This strategy resamples each 3D volume to 128x128x128 and applies subject-wise z-score intensity normalization. This preserves global topological features while standardizing image appearance. Training protocols include mixup, class weighting, and test-time augmentation for robustness. In 5-fold cross-validation, GloResNet achieved 75.18% average accuracy, peaking at 81.82%. It also showed 0.81 specificity and 0.76 sensitivity. This demonstrates its capability as a non-invasive screening tool for neonatal BI.

Key takeaway

For medical imaging researchers developing diagnostic tools for neonatal conditions, this work suggests lightweight 3D CNNs can achieve high predictive accuracy. This is true when combined with topology-preserving preprocessing and robust training techniques. You should consider integrating global manifold mapping and MedicalNet pretraining into your models. This approach offers a non-invasive, efficient screening method for preterm brain injury, potentially accelerating early intervention.

Key insights

Lightweight 3D CNN with global topological features predicts preterm brain injury effectively.

Principles

• Pretraining on MedicalNet addresses data scarcity in medical imaging.
• Global manifold mapping preserves topology in 3D volumes.
• Mixup, class weighting, and TTA enhance model robustness.

Method

Resample 3D MRI volumes to 128x128x128, then apply subject-wise z-score normalization. Train a MedicalNet-pretrained ResNet-10 with mixup, class weighting, and test-time augmentation.

In practice

• Use MedicalNet pretraining for scarce medical imaging data.
• Implement global manifold mapping for 3D volume standardization.
• Apply mixup and test-time augmentation for robust model training.

Topics

• Preterm Brain Injury Prediction
• 3D Convolutional Neural Networks
• Medical Image Analysis
• Global Manifold Mapping
• T2-weighted MRI
• Lightweight CNNs

Code references

• ICL-SUST/GloResNet-Preterm-Brain
• Dstate/UAGLNet
• JuIvyy/GPOcc

Best for: AI Scientist, Research Scientist, Domain Expert

Related on AIssential

• AI for Interoperability in Health Care: Philips’s Carla Goulart Peron — AI in healthcare augments clinicians to expand access and improve diagnostics, with interoperability identified as the most impactful global capability.
• China has approved the world’s first invasive brain-computer chip—here’s what’s next — China's NEO BCI, less invasive and government-backed, is the first globally approved for commercial use, offering hope for paralysis patients.
• ClinEnv: An Interactive Multi-Stage Long Horizon EHR Environment for Agents — ClinEnv reveals LLMs struggle with sequential, uncertain clinical decisions and information gathering in a multi-stage environment.
• Quantitative Movement Testing: Measuring Patient Movements from a Single Smartphone Video — QMT uses smartphone video and deep learning for accessible, accurate 3D kinematic biomarker extraction in clinical settings.
• Predicting the risk of colorectal anastomotic leak based on preoperative mapping of the blood supply of the bowel — An AI-driven system predicts colorectal anastomotic leak risk using CT imaging and deep learning for enhanced surgical planning.
• A Minimalist Brain-Computer Musical Interface for Real-Time Emotion-Driven Sonification: System Design and Preliminary Evaluation — Frontal alpha asymmetry proved unreliable for voluntary emotion control in a real-time brain-computer musical interface.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Takara TLDR - Daily AI Papers.