NA-MIC Project Weeks

Back to Projects List

A deep learning framework for white matter fiber clustering, code release via SlicerDMRI

Key Investigators

• Yuqian Chen (BWH & USYD)
• Chaoyi Zhang (USYD)
• Yang Song (UNSW)
• Tengfei Xue (BWH & USYD)
• Nikos Makris (BWH)
• Yogesh Rathi(BWH)
• Weidong Cai (USYD)
• Fan Zhang (BWH)
• Lauren J O’Donnell (BWH)

Project Description

We propose an unsupervised deep learning framework for fast and effective white matter fiber clustering (WMFC) (Chen et al 2021, MICCAI). It enables parcellation of white matter tractography. Current WMFC methods are facing several challenges such as fiber computation efficiency, sensitivity to fiber direction, combination of spatial and anatomical information, existence of outlier fibers as well as correspondence across subjects. To overcome these challenges, we propose a self-supervised learning strategy to achieve fast and effective WMFC. In this project, we will work on releasing the code of this method. We will provide the trained model and testing samples for demonstration.

Objective

1. Build deep learning training model for white matter fiber clustering and evaluate it on testing data.
2. Code cleaning and releasing.

Approach and Plan

1. In our method, we use a convolutional neural network to learn embeddings of input fibers and improved anatomical coherence of fiber clusters by incorporating brain anatomical information. Outlier removal is performed in a natural way by rejecting fibers with low cluster assignment probability.
2. Experiments are implemented through coding with python.
3. Evaluate our method by performing experiments on three independently acquired datasets.

Progress and Next Steps

1. Code released at: https://github.com/SlicerDMRI/DFC.
2. Trained models and example testing data are provided.

Illustrations

Background and References

Chen, Yuqian, Chaoyi Zhang, Yang Song, Nikos Makris, Yogesh Rathi, Weidong Cai, Fan Zhang, and Lauren J. O’Donnell. “Deep Fiber Clustering: Anatomically Informed Unsupervised Deep Learning for Fast and Effective White Matter Parcellation.” International Conference on Medical Image Computing and Computer-Assisted Intervention. Springer, Cham, 2021.