Related Experiment Video
Updated: Aug 27, 2025
08:49
Whole-Brain Single-Cell Imaging and Analysis of Intact Neonatal Mouse Brains Using MRI, Tissue Clearing, and Light-Sheet Microscopy
Published on: August 1, 2022
3.7K
BrainSec: Automated Brain Tissue Segmentation Pipeline for Scalable Neuropathological Analysis
Zhengfeng Lai1, Luca Cerny Oliveira1, Runlin Guo1
1Department of Electrical and Computer Engineering, University of California Davis, Davis, CA 95616, USA.
IEEE Access : Practical Innovations, Open Solutions
|September 26, 2022
Summary
Automated segmentation of grey and white matter (GM/WM) in brain pathology images using the BrainSec pipeline accelerates neuropathologic deep phenotyping. This CNN-based approach offers robust and scalable analysis for neurodegenerative disease research.
Area of Science:
- Neuroscience
- Computational Pathology
- Medical Image Analysis
Background:
- Neurodegenerative diseases exhibit pathological hallmarks in both grey matter (GM) and white matter (WM).
- Manual segmentation of GM/WM in Whole Slide Images (WSIs) is time-consuming, subjective, and hinders scalable analysis.
- Automating GM/WM segmentation is crucial for deep phenotyping and quantitative pathology in neurodegenerative research.
Purpose of the Study:
- To develop and validate an automated segmentation pipeline (BrainSec) for GM/WM regions in neuropathology WSIs.
- To compare baseline segmentation models (FCN, U-Net) and introduce a robust patch-based CNN approach.
- To integrate automated segmentation with pathology classification for comprehensive disease analysis.
Main Methods:
- A patch-based Convolutional Neural Network (CNN) approach, BrainSec, was developed for GM/WM segmentation.
- Baseline models (FCN, U-Net) were investigated for medical image segmentation.
- A post-processing module was implemented to refine segmentation masks and remove artifacts.
- Gradient-weighted Class Activation Mapping (Grad-CAM) was used for model interpretability.
- BrainSec was integrated with an Amyloid-β pathology classification model.
Main Results:
- BrainSec demonstrated robust and reliable performance across over 180 WSIs, encompassing diverse neuroanatomic regions and cases.
- The pipeline successfully generated XML annotations for visualization in Aperio ImageScope.
- Integration with pathology classification enabled identification, visualization, and quantification of pathologies within segmented GM/WM regions.
- Grad-CAM provided insights into the segmentation process.
Conclusions:
- BrainSec offers an automated, scalable, and robust solution for GM/WM segmentation in neuropathology.
- The pipeline facilitates accurate deep phenotyping and quantitative analysis of neurodegenerative pathologies.
- Integration with pathology classifiers enhances the comprehensive assessment of disease in brain tissue.
