Frontotemporal dementia subtyping using machine learning, multivariate statistics and neuroimaging.

Area of Science:

• Neuroimaging
• Neurology
• Machine Learning

Background:

• Frontotemporal dementia (FTD) is a heterogeneous group of early-onset neurodegenerative disorders.
• Accurate diagnosis of FTD subtypes remains challenging due to overlapping symptoms.
• Magnetic resonance imaging (MRI) is a key tool in supporting FTD diagnosis.

Purpose of the Study:

• To investigate the association between brain atrophy patterns and cognitive impairment severity in FTD.
• To determine if this relationship differs across FTD subtypes.
• To develop a machine learning model for classifying FTD subtypes using neuroimaging and clinical data.

Main Methods:

• Utilized deformation-based morphometry on MRI scans from 136 FTD patients (behavioural variant FTD, semantic variant PPA, non-fluent variant PPA).
• Applied partial least squares (PLS) to assess associations between regional atrophy and cognitive test performance.
• Employed linear regression to analyze group differences in atrophy-cognition relationships and developed a machine learning classifier.

Main Results:

• Four significant latent variables explained 86% of the shared variance between brain atrophy and cognition.
• PLS-based patterns achieved 89.12% cross-validated accuracy in predicting FTD subtypes.
• A clinically feasible model using MRI and two behavioral tests reached 87.18% accuracy, outperforming models using only MRI or behavioral data.

Conclusions:

• The combination of brain atrophy and clinical characteristics, analyzed with multivariate statistics, serves as a biomarker for FTD phenotyping.
• Deformation-based morphometry measures enhance classification accuracy, even with limited clinical testing.
• Integrating MRI and clinical data offers a powerful approach for precise FTD subtype classification.