GloResNet: A lightweight 3D CNN with global topological features for preterm brain injury prediction
Researchers have developed GloResNet, a lightweight 3D convolutional neural network designed to predict brain injury in preterm infants using T2-weighted MRI scans. The model, based on ResNet-10 and pretrained on MedicalNet, incorporates a global manifold mapping strategy to preserve topological features while standardizing image appearance. In cross-validation tests, GloResNet achieved an average accuracy of 75.18%, demonstrating its potential as a non-invasive screening tool for neonatal brain injury. AI
IMPACT Offers a potential new non-invasive tool for early detection of brain injury in newborns.