Artificial Intelligence Based Hierarchical Classification of Frontotemporal Dementia.

Area of Science:

• Neuroimaging
• Machine Learning
• Neurology

Background:

• Frontotemporal dementia (FTD) is a presenile dementia with distinct subtypes: behavioral-variant FTD (bvFTD), non-fluent variant primary progressive aphasia (nfvPPA), and semantic variant primary progressive aphasia (svPPA).
• Accurate classification of FTD subtypes is crucial for understanding disease heterogeneity and developing targeted interventions.

Purpose of the Study:

• To develop and validate a data-driven hierarchical classification model for brain images across FTD subtypes.
• To compare the hierarchical model's performance against traditional flat multi-class models and other machine learning approaches.

Main Methods:

• Utilized MRI-derived cortical and subcortical measurements from 300 subjects (bvFTD, svPPA, nfvPPA, Alzheimer's Disease, cognitively normal) from the FTD Neuroimaging Initiative.
• Applied hierarchical classification using Support Vector Machine (SVM), Linear Discriminant Analysis, and Naive Bayes algorithms.
• Compared results with flat multi-class models, Convolutional Neural Networks (CNNs), and ensemble multi-layer perception models.

Main Results:

• The hierarchical model achieved high classification accuracies: 87.42% (SVM), 83.23% (LDA), and 82.44% (Naive Bayes) for five classes.
• Demonstrated significant improvement over flat multi-class models, which yielded lower accuracies (e.g., 82.80% for SVM).
• Hierarchical approach performed comparably to CNNs and outperformed state-of-the-art methods in a three-class (CN, Non-FTD, FTD) comparison.

Conclusions:

• The hierarchical classification approach effectively captures the heterogeneity within FTD subtypes.
• This method offers potential for improved diagnostic accuracy and personalized treatment strategies for FTD patients.
• Findings support the clinical relevance of understanding FTD subtype-specific characteristics for tailored care.