Regional CSF volume quantification using deep learning for comparative analysis of brain atrophy in frontotemporal...

Area of Science:

• Neuroimaging
• Artificial Intelligence
• Neurology

Background:

• Frontotemporal dementia (FTD) presents diverse clinical symptoms, making subtype differentiation via imaging difficult.
• Accurate diagnosis of FTD subtypes is crucial for effective patient management and treatment strategies.

Purpose of the Study:

• To develop and validate a deep learning model for quantifying cerebrospinal fluid (CSF) volumes to measure brain atrophy.
• To assess the model's ability to differentiate between FTD subtypes (bvFTD, nfvPPA, svPPA), Alzheimer's disease (DAT), and cognitively unimpaired (CU) individuals.

Main Methods:

• A deep learning model was trained on 3D T1-weighted MRI scans from 1,854 participants (CU, DAT, bvFTD, nfvPPA, svPPA).
• The model quantified CSF volumes in 14 regions of interest (RoIs) and generated age- and sex-adjusted W-scores to indicate regional atrophy.
• Standard MRI scans with minimal preprocessing were utilized.

Main Results:

• Distinct, lateralized atrophy patterns were identified for each FTD subtype.
• Behavioral variant FTD (bvFTD) showed bilateral frontal and right-predominant parietal/temporal atrophy.
• Nonfluent variant primary progressive aphasia (nfvPPA) displayed left-predominant frontal/parietal atrophy.
• Semantic variant PPA (svPPA) exhibited marked left-lateralized temporal and hippocampal atrophy.
• All FTD subtypes showed significantly greater CSF expansion in characteristic regions compared to DAT and CU controls.

Conclusions:

• The deep learning model provides a simple, interpretable measure of brain atrophy.
• This approach effectively differentiates FTD subtypes using standard MRI data.
• The method demonstrates clinical utility for diagnosing and managing FTD.