Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

DDEvENet: Evidence-based Ensemble Learning for Uncertainty-aware Brain Parcellation Using Diffusion MRI

Chenjun Li, Dian Yang, Shun Yao, Shuyue Wang, Ye Wu, Le Zhang, Qiannuo Li, Kang Ik Kevin Cho, Johanna Seitz-Holland, Lipeng Ning, Jon Haitz Legarreta, Yogesh Rathi, Carl-Fredrik Westin, Lauren J. O'Donnell, Nir A. Sochen, Ofer Pasternak, Fan Zhang

TL;DR

This work introduces EVENet, an evidential deep learning–driven ensemble for directly parcellating the brain from diffusion MRI while delivering voxel-wise uncertainty maps. By employing five parameter-specific subnetworks (FA, MD, E1, E2, E3) and an evidence-based fusion strategy, it achieves state-of-the-art parcellation accuracy across diverse datasets and acquisition protocols. The approach also provides clinically meaningful uncertainty that highlights boundaries and pathological regions, improving interpretability and potential lesion detection. Its generalizability, efficiency, and compatibility with existing pipelines position EVENet as a robust tool for large-scale neuroimaging studies and clinical applications.

Abstract

In this study, we developed an Evidence-based Ensemble Neural Network, namely EVENet, for anatomical brain parcellation using diffusion MRI. The key innovation of EVENet is the design of an evidential deep learning framework to quantify predictive uncertainty at each voxel during a single inference. To do so, we design an evidence-based ensemble learning framework for uncertainty-aware parcellation to leverage the multiple dMRI parameters derived from diffusion MRI. Using EVENet, we obtained accurate parcellation and uncertainty estimates across different datasets from healthy and clinical populations and with different imaging acquisitions. The overall network includes five parallel subnetworks, where each is dedicated to learning the FreeSurfer parcellation for a certain diffusion MRI parameter. An evidence-based ensemble methodology is then proposed to fuse the individual outputs. We perform experimental evaluations on large-scale datasets from multiple imaging sources, including high-quality diffusion MRI data from healthy adults and clinically diffusion MRI data from participants with various brain diseases (schizophrenia, bipolar disorder, attention-deficit/hyperactivity disorder, Parkinson's disease, cerebral small vessel disease, and neurosurgical patients with brain tumors). Compared to several state-of-the-art methods, our experimental results demonstrate highly improved parcellation accuracy across the multiple testing datasets despite the differences in dMRI acquisition protocols and health conditions. Furthermore, thanks to the uncertainty estimation, our EVENet approach demonstrates a good ability to detect abnormal brain regions in patients with lesions, enhancing the interpretability and reliability of the segmentation results.

DDEvENet: Evidence-based Ensemble Learning for Uncertainty-aware Brain Parcellation Using Diffusion MRI

TL;DR

This work introduces EVENet, an evidential deep learning–driven ensemble for directly parcellating the brain from diffusion MRI while delivering voxel-wise uncertainty maps. By employing five parameter-specific subnetworks (FA, MD, E1, E2, E3) and an evidence-based fusion strategy, it achieves state-of-the-art parcellation accuracy across diverse datasets and acquisition protocols. The approach also provides clinically meaningful uncertainty that highlights boundaries and pathological regions, improving interpretability and potential lesion detection. Its generalizability, efficiency, and compatibility with existing pipelines position EVENet as a robust tool for large-scale neuroimaging studies and clinical applications.

Abstract

In this study, we developed an Evidence-based Ensemble Neural Network, namely EVENet, for anatomical brain parcellation using diffusion MRI. The key innovation of EVENet is the design of an evidential deep learning framework to quantify predictive uncertainty at each voxel during a single inference. To do so, we design an evidence-based ensemble learning framework for uncertainty-aware parcellation to leverage the multiple dMRI parameters derived from diffusion MRI. Using EVENet, we obtained accurate parcellation and uncertainty estimates across different datasets from healthy and clinical populations and with different imaging acquisitions. The overall network includes five parallel subnetworks, where each is dedicated to learning the FreeSurfer parcellation for a certain diffusion MRI parameter. An evidence-based ensemble methodology is then proposed to fuse the individual outputs. We perform experimental evaluations on large-scale datasets from multiple imaging sources, including high-quality diffusion MRI data from healthy adults and clinically diffusion MRI data from participants with various brain diseases (schizophrenia, bipolar disorder, attention-deficit/hyperactivity disorder, Parkinson's disease, cerebral small vessel disease, and neurosurgical patients with brain tumors). Compared to several state-of-the-art methods, our experimental results demonstrate highly improved parcellation accuracy across the multiple testing datasets despite the differences in dMRI acquisition protocols and health conditions. Furthermore, thanks to the uncertainty estimation, our EVENet approach demonstrates a good ability to detect abnormal brain regions in patients with lesions, enhancing the interpretability and reliability of the segmentation results.
Paper Structure (19 sections, 11 equations, 5 figures, 4 tables)

This paper contains 19 sections, 11 equations, 5 figures, 4 tables.

Table of Contents

  1. Introduction
  2. Methods
  3. Network Architecture
  4. Evidential Deep Learning
  5. Evidence-based Ensemble Learning for Uncertainty Estimation
  6. Implementation and Parameter Settings
  7. Experimental Evaluation
  8. Experimental Datasets
  9. MRI Acquisition and Preprocessing
  10. Experimental Design
  11. Ablation Study
  12. Comparison to State-of-the-art Methods
  13. Comparison of Parcellation Performance on Different Datasets
  14. Experimental Results
  15. Ablation Study
  16. ...and 4 more sections

Figures (5)

  • Figure 1: Example of out-of-distribution parcellation. (a) shows the input image data with apparent abnormal lesion regions (glioma) within the red rectangle, which constitutes this scan as an out-of-distribution scan. (b) shows the parcellation result using the widely used FreeSurfer software fischl2012freesurfer, where the lesion boundary is mislabeled as being part of the cortex. (c) shows the parcellation result using FastSurfer henschel2020fastsurfer, a recently proposed deep-learning approach. While we can observe a visually improved parcellation, there is still apparent mislabeling in the glioma (shown with yellow arrows). (d) shows the uncertainty estimation using our proposed EVENet method, which outlines tissue boundaries as well as abnormal brain regions and is utilized to improve parcellation.
  • Figure 2: EVENet overview. Five parameter maps calculated from the diffusion-weighted images (DWIs) are used as input to train the corresponding subnetworks, and the FreeSurfer-based parcellation is used as the ground truth. Incorporating evidential loss, the subnetworks produce voxel-wise evidence that can be further parameterized as Dirichlet distributions and output evidence-based uncertainty. The evidence-based uncertainty is used as a criterion for ensemble prediction maps, and the final uncertainty heatmap is calculated using the entropy of the averaged evidence.
  • Figure 3: Visualization of parcellation results on randomly selected PPMI (left) and CNP (right) scans. The green and yellow arrows identify examples of tissue boundaries for easier comparison across parcellation labelmaps.
  • Figure 4: Visualization of uncertainty estimations on a randomly selected patient scan with WMH. The left part shows the input and the evidence-based uncertainty from three of the subnetworks with a comparison to the manually segmented WMH mask. The right side shows the final parcellation and uncertainty estimation.
  • Figure 5: Visualization of uncertainty estimations on another randomly selected patient scan with BT. This scan fails to run properly on the FS software due to the existence of the tumor region. The left part shows the input and the evidence-based uncertainty from three of the subnetworks. The final output with a comparison to the manually segmented mask is provided.