liamchalcroft/midl-2025-base

SimCLR-MRI Pre-trained Encoder (Base)

This repository contains a pre-trained 3D CNN encoder for MRI analysis. The model was trained using contrastive learning (SimCLR) on MPRAGE brain MRI scans, using standard image augmentations.

Model Description

The encoder is a 3D CNN with 5 convolutional blocks (64, 128, 256, 512, 768 channels), outputting 768-dimensional features. This base variant was trained on real MPRAGE scans using standard contrastive augmentations (random rotations, flips, intensity changes).

Training Procedure

• Pre-training Data: 51 qMRI datasets (22 healthy, 29 stroke subjects)
• Augmentations: Standard geometric and intensity transformations
• Input: 3D MPRAGE volumes (96×96×96)
• Output: 768-dimensional feature vectors

Intended Uses

This encoder is particularly suited for:

• Transfer learning on T1-weighted MRI tasks
• Feature extraction for structural MRI analysis
• General brain MRI representation learning