Artificial Intelligence Based Hierarchical Classification of Frontotemporal Dementia.
Journal:
Annual International Conference of the IEEE Engineering in Medicine and Biology Society. IEEE Engineering in Medicine and Biology Society. Annual International Conference
PMID:
40039638
Abstract
Frontotemporal dementia (FTD) is a typical kind of presenile dementia with three main subtypes: behavioral-variant FTD (bvFTD), non-fluent variant primary progressive aphasia (nfvPPA), and semantic variant primary progressive aphasia (svPPA). Our aim is to classify brain images of each subject into one of the spectrums of the FTD in a hierarchical order by applying data-driven techniques. Specifically, we took 300 subjects (30 bvFTD, 41 svPPA, 25 nfvPPA, 80 Alzheimer's Disease, and 124 cognitively normal) from the Frontotemporal Lobar Degeneration Neuroimaging Initiative archive to validate our approach. The cortical and subcortical measurements, e.g., cortical thickness and volumetric segmentation, were extracted from MRI images for the experiment. Our proposed model yielded classification accuracy of 87.42%, 83.23%, and 82.44% with Support Vector Machine (SVM), Linear Discriminant Analysis, and Naive Bayes methods, respectively, for five classes. We observed a significant improvement over the flat multi-class model, which has an accuracy of 82.80%, 81.27%, and 80.45%, respectively, for five classes. We also compared our proposed approach with a Convolutional Neural Networks model and ensemble with multi-layer perception as well. The results showed that the hierarchical approach performs almost equally well here as in the SVM model. Moreover, we also tested our hierarchical approach for three classes (CN, Non-FTD, FTD) to compare with state-of-the-art methods and obtained better results.Clinical relevance- This study provides a detailed understanding of the heterogeneity within the FTD subtypes; clinicians can better tailor diagnostic and treatment strategies to match each subtype's specific characteristics and progression patterns.