Anatomical Foundation Models for Brain MRIs

TL;DR

This work introduces AnatCL, a brain MRI foundation framework that augments weakly supervised contrastive learning with anatomical descriptors (CT, GMV, SA) alongside age to learn robust, transferable representations. Pretrained on OpenBHB, AnatCL is validated across 12 downstream tasks and 10 clinical phenotypes using multiple public datasets, consistently outperforming age-only and other contrastive baselines. The approach defines local and global descriptor variants to capture regional and holistic anatomical information, and shows that anatomy-informed pretraining yields more generalizable embeddings than traditional brain-age pretraining. The authors also report ablations on feature choices and model architecture, and release pretrained weights to enable broader adoption in neuroimaging analyses and clinical prediction tasks.

Abstract

Deep Learning (DL) in neuroimaging has become increasingly relevant for detecting neurological conditions and neurodegenerative disorders. One of the most predominant biomarkers in neuroimaging is represented by brain age, which has been shown to be a good indicator for different conditions, such as Alzheimer's Disease. Using brain age for weakly supervised pre-training of DL models in transfer learning settings has also recently shown promising results, especially when dealing with data scarcity of different conditions. On the other hand, anatomical information of brain MRIs (e.g. cortical thickness) can provide important information for learning good representations that can be transferred to many downstream tasks. In this work, we propose AnatCL, an anatomical foundation model for brain MRIs that i.) leverages anatomical information in a weakly contrastive learning approach, and ii.) achieves state-of-the-art performances across many different downstream tasks. To validate our approach we consider 12 different downstream tasks for the diagnosis of different conditions such as Alzheimer's Disease, autism spectrum disorder, and schizophrenia. Furthermore, we also target the prediction of 10 different clinical assessment scores using structural MRI data. Our findings show that incorporating anatomical information during pre-training leads to more robust and generalizable representations. Pre-trained models can be found at: https://github.com/EIDOSLAB/AnatCL.

Figures (3)

• Figure 1: We propose AnatCL, a contrastive learning approach that leverages patient's metadata (i.e., age) alongside anatomical information derived from structural MRIs, in order to learn meaningful and general representations. Given two input samples $x$ and $x_i$, the objective of AnatCL is to learn embeddings $z$ and $z_i$, such that their distance in the representation space is proportional to the difference in age and brain anatomy between the two samples.
• Figure 2: Graphical evaluation of AnatCL (local and global) on all deep phenotyping tasks considered. Overall, AnatCL achieves higher results more consistently than any other baseline, with notable performance on ASD and Schizophrenia.
• Figure 3: Prediction of clinical assessment scores from structural brain MRIs.