New deep learning model sets accuracy benchmark for extracting infant brains from MRI scans
-
A new deep learning model called ANUBEX is proposed for automatically extracting the brain from neonatal MRI scans. It is based on the nnU-Net architecture and trained on a large, multi-institutional dataset.
-
ANUBEX demonstrated the highest accuracy in extracting the brain compared to 5 other publicly available methods when tested on an independent dataset of neonatal MRIs from multiple institutions.
-
The model performs well on both 2D and 3D MRI scans with varying resolutions and acquisition parameters thanks to the heterogeneous training data.
-
Additional analyses showed similar performance on T1- and T2-weighted scans, good accuracy on preterm infant scans, and robustness to motion artifacts.
-
By sharing the trained model weights publicly, the authors aim to provide an off-the-shelf tool to standardize the neonatal brain extraction step across institutions and studies.