Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

wmh_seg: Transformer based U-Net for Robust and Automatic White Matter Hyperintensity Segmentation across 1.5T, 3T and 7T

Jinghang Li, Tales Santini, Yuanzhe Huang, Joseph M. Mettenburg, Tamer S. Ibrahim, Howard J. Aizenstein, Minjie Wu

TL;DR

This work introduces wmh_seg, a novel deep learning model leveraging a transformer-based encoder from SegFormer that is the first that offers quality white matter lesion segmentation on 7T FLAIR images.

Abstract

White matter hyperintensity (WMH) remains the top imaging biomarker for neurodegenerative diseases. Robust and accurate segmentation of WMH holds paramount significance for neuroimaging studies. The growing shift from 3T to 7T MRI necessitates robust tools for harmonized segmentation across field strengths and artifacts. Recent deep learning models exhibit promise in WMH segmentation but still face challenges, including diverse training data representation and limited analysis of MRI artifacts' impact. To address these, we introduce wmh_seg, a novel deep learning model leveraging a transformer-based encoder from SegFormer. wmh_seg is trained on an unmatched dataset, including 1.5T, 3T, and 7T FLAIR images from various sources, alongside with artificially added MR artifacts. Our approach bridges gaps in training diversity and artifact analysis. Our model demonstrated stable performance across magnetic field strengths, scanner manufacturers, and common MR imaging artifacts. Despite the unique inhomogeneity artifacts on ultra-high field MR images, our model still offers robust and stable segmentation on 7T FLAIR images. Our model, to date, is the first that offers quality white matter lesion segmentation on 7T FLAIR images.

wmh_seg: Transformer based U-Net for Robust and Automatic White Matter Hyperintensity Segmentation across 1.5T, 3T and 7T

TL;DR

This work introduces wmh_seg, a novel deep learning model leveraging a transformer-based encoder from SegFormer that is the first that offers quality white matter lesion segmentation on 7T FLAIR images.

Abstract

White matter hyperintensity (WMH) remains the top imaging biomarker for neurodegenerative diseases. Robust and accurate segmentation of WMH holds paramount significance for neuroimaging studies. The growing shift from 3T to 7T MRI necessitates robust tools for harmonized segmentation across field strengths and artifacts. Recent deep learning models exhibit promise in WMH segmentation but still face challenges, including diverse training data representation and limited analysis of MRI artifacts' impact. To address these, we introduce wmh_seg, a novel deep learning model leveraging a transformer-based encoder from SegFormer. wmh_seg is trained on an unmatched dataset, including 1.5T, 3T, and 7T FLAIR images from various sources, alongside with artificially added MR artifacts. Our approach bridges gaps in training diversity and artifact analysis. Our model demonstrated stable performance across magnetic field strengths, scanner manufacturers, and common MR imaging artifacts. Despite the unique inhomogeneity artifacts on ultra-high field MR images, our model still offers robust and stable segmentation on 7T FLAIR images. Our model, to date, is the first that offers quality white matter lesion segmentation on 7T FLAIR images.
Paper Structure (15 sections, 3 equations, 9 figures, 1 table)

This paper contains 15 sections, 3 equations, 9 figures, 1 table.

Table of Contents

  1. Introduction
  2. Related Work
  3. Classical White Matter Lesion Segmentation
  4. Deep Learning Approaches
  5. Contribution
  6. Material and Methods
  7. Data Distribution
  8. Data Augmentation
  9. Network Architecture
  10. Implementation
  11. Results
  12. Discussion
  13. Conclusion
  14. Author Contribution
  15. Acknowledgment

Figures (9)

  • Figure 1: Axial slices of T2 weighted FLAIR images acquired different institutes. These axial slices illustrate the significant MR image variability due to the differences in MR scanner manufacturer, magnetic field strength, and sequence acquisition protocol.
  • Figure 2: Commonly observed MRI artifacts implemented for data augmentation using torchio perez2021torchio
  • Figure 3: wmh_seg model architecture. The model consists of hierarchical transformer encoder and convolutional decoder.
  • Figure 4: wmh_seg and pgs segmentation results on 1.5 Tesla, 3 Tesla FLAIR images.
  • Figure 5: wmh_seg and pgs segmentation results on 7 Tesla FLAIR images.
  • ...and 4 more figures