Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Subthalamic Nucleus segmentation in high-field Magnetic Resonance data. Is space normalization by template co-registration necessary?

Tomás Lima, Igor Varga, Eduard Bakštein, Daniel Novák, Victor Alves

TL;DR

The paper investigates STN segmentation for DBS planning in Parkinson's disease using high-field 7T MRI and compares two deep learning pipelines: a template-based approach with coarse space normalization to MNI space and a native-space segmentation approach. Using nnUNet on three public datasets, the native-space method outperforms the template-based method for STN while delivering comparable results for Red Nucleus and Substantia Nigra. The template-based approach suffers from boundary artefacts after inverse transformation, whereas the native-space method benefits from more aggressive data augmentation and avoids template-induced distortions. The findings suggest that template normalization is not strictly necessary for high-quality STN segmentation and point toward broader applicability to lower-field MRI with appropriate preprocessing and augmentation strategies.

Abstract

Deep Brain Stimulation (DBS) is one of the most successful methods to diminish late-stage Parkinson's Disease (PD) symptoms. It is a delicate surgical procedure which requires detailed pre-surgical patient's study. High-field Magnetic Resonance Imaging (MRI) has proven its improved capacity of capturing the Subthalamic Nucleus (STN) - the main target of DBS in PD - in greater detail than lower field images. Here, we present a comparison between the performance of two different Deep Learning (DL) automatic segmentation architectures, one based in the registration to a brain template and the other performing the segmentation in in the MRI acquisition native space. The study was based on publicly available high-field 7 Tesla (T) brain MRI datasets of T1-weighted and T2-weighted sequences. nnUNet was used on the segmentation step of both architectures, while the data pre and post-processing pipelines diverged. The evaluation metrics showed that the performance of the segmentation directly in the native space yielded better results for the STN segmentation, despite not showing any advantage over the template-based method for the to other analysed structures: the Red Nucleus (RN) and the Substantia Nigra (SN).

Subthalamic Nucleus segmentation in high-field Magnetic Resonance data. Is space normalization by template co-registration necessary?

TL;DR

The paper investigates STN segmentation for DBS planning in Parkinson's disease using high-field 7T MRI and compares two deep learning pipelines: a template-based approach with coarse space normalization to MNI space and a native-space segmentation approach. Using nnUNet on three public datasets, the native-space method outperforms the template-based method for STN while delivering comparable results for Red Nucleus and Substantia Nigra. The template-based approach suffers from boundary artefacts after inverse transformation, whereas the native-space method benefits from more aggressive data augmentation and avoids template-induced distortions. The findings suggest that template normalization is not strictly necessary for high-quality STN segmentation and point toward broader applicability to lower-field MRI with appropriate preprocessing and augmentation strategies.

Abstract

Deep Brain Stimulation (DBS) is one of the most successful methods to diminish late-stage Parkinson's Disease (PD) symptoms. It is a delicate surgical procedure which requires detailed pre-surgical patient's study. High-field Magnetic Resonance Imaging (MRI) has proven its improved capacity of capturing the Subthalamic Nucleus (STN) - the main target of DBS in PD - in greater detail than lower field images. Here, we present a comparison between the performance of two different Deep Learning (DL) automatic segmentation architectures, one based in the registration to a brain template and the other performing the segmentation in in the MRI acquisition native space. The study was based on publicly available high-field 7 Tesla (T) brain MRI datasets of T1-weighted and T2-weighted sequences. nnUNet was used on the segmentation step of both architectures, while the data pre and post-processing pipelines diverged. The evaluation metrics showed that the performance of the segmentation directly in the native space yielded better results for the STN segmentation, despite not showing any advantage over the template-based method for the to other analysed structures: the Red Nucleus (RN) and the Substantia Nigra (SN).
Paper Structure (11 sections, 3 equations, 8 figures, 4 tables)

This paper contains 11 sections, 3 equations, 8 figures, 4 tables.

Table of Contents

  1. Introduction
  2. Materials and Methods
  3. MRI Data
  4. Training and Test Subsets
  5. Labels Generation
  6. Template-Based Automatic Segmentation - Method I
  7. Native Space Automatic Segmentation - Method II
  8. Evaluation of the Automatic Segmentations
  9. Results
  10. Discussion
  11. Conclusions

Figures (8)

  • Figure 1: Example of a segmentation map output by the pBrain pipeline overlayed on top of the corresponding T2w image.
  • Figure 2: Overall pre-processing pipeline of the MNI based automatic segmentation method.
  • Figure 3: Overall pre-processing pipeline of Method II architecture.
  • Figure 4: Example of what the CoM of a segmentation map represents in views of the axial and coronal planes on a T2w image with overlay of a segmentation map with the six different labels.
  • Figure 5: Localizer+Segmenter architecture workflow diagram after data pre-processing.
  • ...and 3 more figures