Detecting Alzheimer's Disease Stages and Frontotemporal Dementia in Time Courses of Resting-State fMRI Data Using a...

Area of Science:

• Neuroscience
• Medical Imaging
• Artificial Intelligence

Background:

• Distinguishing between neurodegenerative dementia subtypes like Alzheimer's disease (AD) and frontotemporal dementia (FTD) is clinically challenging due to overlapping symptoms and atypical presentations.
• Resting-state functional magnetic resonance imaging (rs-fMRI) reveals condition-specific brain alterations in AD, FTD, and mild cognitive impairment (MCI) compared to healthy controls (HC).

Purpose of the Study:

• To develop and evaluate a machine learning model for accurate classification of AD, FTD, healthy controls (HC), and MCI using rs-fMRI data.
• To assess the impact of integrating clinical variables with imaging data on diagnostic model performance.

Main Methods:

• rs-fMRI data and clinical information were curated from the ADNI and FTLDNI databases.
• Preprocessing, feature extraction, and gradient-boosted decision trees with nested cross-validation were employed to build classification models.
• Model performance was evaluated using balanced accuracy, macro-averaged AUC, and macro-averaged F1 scores on unseen test data.

Main Results:

• A model using only rs-fMRI features achieved 74.4% balanced accuracy, misclassifying AD scans as MCI.
• Integrating clinical variables significantly improved model performance, yielding 91.1% balanced accuracy, 0.99 AUC, and 0.92 F1 score.
• The multimodal model demonstrated high accuracy in classifying FTD (F1=0.99), HC (F1=0.99), MCI (F1=0.86), and notably improved AD classification (F1=0.74).

Conclusions:

• A multimodal machine learning approach combining rs-fMRI and clinical data offers a robust and accurate method for differentiating AD, FTD, MCI, and HC.
• This approach holds significant potential for improving early and precise diagnosis of dementia subtypes, guiding treatment strategies.