Better Generalization of White Matter Tract Segmentation to Arbitrary Datasets with Scaled Residual Bootstrap

Abstract

White matter (WM) tract segmentation is a crucial step for brain connectivity studies. It is performed on diffusion magnetic resonance imaging (dMRI), and deep neural networks (DNNs) have achieved promising segmentation accuracy. Existing DNN-based methods use an annotated dataset for model training. However, the performance of the trained model on a different test dataset may not be optimal due to distribution shift, and it is desirable to design WM tract segmentation approaches that allow better generalization of the segmentation model to arbitrary test datasets. In this work, we propose a WM tract segmentation approach that improves the generalization with scaled residual bootstrap. The difference between dMRI scans in training and test datasets is most noticeably caused by the different numbers of diffusion gradients and noise levels. Since both of them lead to different signal-to-noise ratios (SNRs) between the training and test data, we propose to augment the training scans by adjusting the noise magnitude and develop an adapted residual bootstrap strategy for the augmentation. First, with a dictionary-based linear representation of diffusion signals, we compute the signal residuals for the training dMRI scans, which can represent samples drawn from the noise distribution. Then, we adapt the bootstrap procedure by scaling the residuals that are randomly drawn with replacement and adding the scaled residuals to the linear signal representation, where augmented dMRI scans with different SNRs are generated. Finally, the augmented and original images are jointly included in model training. Since it is difficult to know the SNR of the test data a priori, we choose to perform the residual scaling with multiple factors. To validate the proposed approach, two dMRI datasets were used, and the experimental results show that our method consistently improved the generalization of WM tract segmentation under various settings.