Automated deep learning segmentation of high-resolution 7 T postmortem MRI for quantitative analysis of structure-pathology correlations in neurodegenerative diseases (2303.12237v2)

Abstract: Postmortem MRI allows brain anatomy to be examined at high resolution and to link pathology measures with morphometric measurements. However, automated segmentation methods for brain mapping in postmortem MRI are not well developed, primarily due to limited availability of labeled datasets, and heterogeneity in scanner hardware and acquisition protocols. In this work, we present a high resolution of 135 postmortem human brain tissue specimens imaged at 0.3 mm$^{3}$ isotropic using a T2w sequence on a 7T whole-body MRI scanner. We developed a deep learning pipeline to segment the cortical mantle by benchmarking the performance of nine deep neural architectures, followed by post-hoc topological correction. We then segment four subcortical structures (caudate, putamen, globus pallidus, and thalamus), white matter hyperintensities, and the normal appearing white matter. We show generalizing capabilities across whole brain hemispheres in different specimens, and also on unseen images acquired at 0.28 mm^3 and 0.16 mm^3 isotropic T2*w FLASH sequence at 7T. We then compute localized cortical thickness and volumetric measurements across key regions, and link them with semi-quantitative neuropathological ratings. Our code, Jupyter notebooks, and the containerized executables are publicly available at: https://pulkit-khandelwal.github.io/exvivo-brain-upenn

• The paper introduces a deep learning segmentation framework that accurately delineates brain structures in high-resolution 7T postmortem MRI.
• It employs the nnU-Net model with post-hoc topological corrections to ensure robust morphometric accuracy across diverse neurodegenerative pathologies.
• Cortical thickness measurements from the segmentation strongly correlate with neuropathological ratings, highlighting its potential for advancing neurodegenerative research.

Automated Deep Learning Segmentation of High-Resolution 7 T Postmortem MRI for Neurodegenerative Disease Research

This paper presents an automated deep learning framework for segmenting high-resolution postmortem brain MRI, conducted at 7T. It addresses a critical need in the analysis of postmortem MRI, which provides detailed brain anatomy and can facilitate correlations between neurodegenerative pathological processes and structural changes in the brain. The paper evaluates nine different neural network architectures for accurately segmenting key brain structures on 135 whole brain hemispheres with diverse neurodegenerative pathologies.

The authors employ a high-resolution T2-weighted MRI sequence, capturing images at 0.3 mm isotropic resolution, which allows for detailed examination of neuroanatomical structures. The deep learning pipeline uses an nnU-Net framework, which showed robust performance in segmenting cortical gray matter, subcortical structures, white matter, and WM hyperintensities. Notably, the nnU-Net model demonstrated superior generalization capabilities across various MRI sequences and unseen imaging protocols, suggesting its utility in diverse datasets of postmortem MRI.

A significant contribution of this research lies in the post-hoc topological corrections applied to the segmentations, ensuring geometric and morphological accuracy, essential for deriving reliable morphometric measurements. Moreover, cortical thickness measurements derived from the automated segmentations exhibited high correlations with reference manual annotations, validating the pipeline's reliability.

The paper further investigates associations between regional cortical thickness and semi-quantitative neuropathological ratings, such as regional p-tau and neuronal loss, and amyloid-beta, CERAD, and Braak staging. The significant negative correlations found, particularly in brain regions associated with Alzheimer's disease pathology, emphasize the pipeline's potential in elucidating structure-pathology relationships in neurodegenerative diseases.

While the paper advances the field of postmortem MRI analysis, there are limitations, including the relatively small training data for whole hemisphere segmentations. Future work entails expanding the dataset, incorporating quantitative image registration methods, and verifying the generalizability of the approach to other neurodegenerative disease populations and imaging modalities. Furthermore, addressing tissue changes due to post-mortem and fixation processes remains crucial for improving antemortem and postmortem matching.

In summary, this work represents a rigorous evaluation of deep learning methods for postmortem MRI analysis. It positions automated segmentation as a viable tool for comprehensive neuroanatomical studies, aiding the development of antemortem biomarkers for neurodegenerative diseases. The release of the segmentation pipeline and the shared findings aim to catalyze further research, with broad implications for combining imaging modalities and neuropathological assessments to advance our understanding of neurodegeneration.