[CVPR2024] Fully convolutional slice-to-volume reconstruction for single-stack MRI

This is the official repo for the CVPR 2024 paper "Fully Convolutional Slice-to-Volume Reconstruction (FC-SVR) for Single-Stack MRI" by Sean I Young, Yaël Balbastre, Bruce Fischl and others.

Pre-requisites

All pre-requisite python packages are listed in pytorch_1.13.1.yml. Run conda env create -f pytorch_1.13.1.yml. Also, FreeSurfer version 7 is required to prepare the training dataset. See https://surfer.nmr.mgh.harvard.edu/fswiki/rel7downloads

Dataset Preparation

The Havard CRL fetal atlases can be downloaded from http://crl.med.harvard.edu/research/fetal_brain_atlas. Preprocess the data using preprocess/crl.py
The FeTA training and validation volumes can be downloaded from https://doi.org/10.7303/syn25649159. Preprocess the data using preprocess/feta.py

Training

Run feta3d_svr_train.sh to train the svr model on the FeTA 2.1 data. Run feta3d_inpaint_train.sh to train the interpolation model.

Inference

Reconstructions on adult brain data

Reconstructions on the FeTA dataset

Pretrained Weights

The pretrained weights for motion estimation and interpolation networks are available from: https://drive.google.com/drive/folders/1LSuhLNwTzZtonifYVjyPtZ1si-I2RwXB.

Work with us!

Feel free to reach out to me via e-mail and say hello if you have interesting ideas for extensions, applications or if you simply just want to chat!