liamchalcroft/midl-2025-seq-aug

SimCLR-MRI Pre-trained Encoder (SeqAug)

This repository contains a pre-trained 3D CNN encoder for MRI analysis. The model was trained using contrastive learning (SimCLR) with Bloch equation simulations to generate multi-contrast views of the same anatomy.

Model Description

The encoder is a 3D CNN with 5 convolutional blocks (64, 128, 256, 512, 768 channels), outputting 768-dimensional features. This SeqAug variant treats different simulated MRI sequences as strong augmentations during contrastive learning, encouraging sequence-robust representations.

Training Procedure

• Pre-training Data: 51 qMRI datasets (22 healthy, 29 stroke subjects)
• Augmentations: Bloch simulation-based sequence augmentation + standard transformations
• Input: 3D MRI volumes (96×96×96)
• Output: 768-dimensional feature vectors

Intended Uses

This encoder is particularly suited for:

• Cross-sequence transfer learning
• Multi-contrast MRI analysis
• Sequence-robust feature extraction