Contribution to the BraTS-Path 2024 Challenge organized during MICCAI 2024 conference.
The repository contains the source code used to prepare the submission to the 1st edition of the BraTS-Path Challenge.
The submission scored the 1st place in the competition - outperforming the fine-tuned general-purpose computer vision models.
- Download the ProvGigaPath pretrained model ProvGigaPath. You need to download only the patch-level encoder. The slide-level encoder is not used. Remember about giving credit to the ProvGigaPath authors! Click.
- Navigate to the Paths and set the path to the BraTS-Path dataset and the ProvGigaPath model.
- Parse the BraTS-Path dataset (available upon request from the challenge organizers): Parse.
- Modify the training parameters to start the training without any previous checkpoints.
- Wait until the training finishes. Then you can use the evaluation scripts to test the performance.
- Register to the BraTS-Path Challenge and follow the guidelines there related to the MLCube preparation.
The source code is released under Apache-2.0 license. However, note that the source code for ProvGigaPath and the associated pretrained model follow a diferent license.