Investigation of multimodal deep-learning architectures for the classification of Alzheimer's disease and amyloid status

Alzheimer’s disease is a progressive neuro-degenerative disorder that affects global health. The amyloid beta protein (Aβ) is an important biomarker of Alzheimer’s disease, with levels increasing early in the course of the disease. Patients can be categorized into normal and abnormal in terms of amyloid status. Since currently measuring Aβ requires invasive procedures, we present a method to predict amyloid status using multi-modal medical image data with transformer-based machine learning architectures. Data from ADNI included structural T1-weighted MRI and FDG-PET scans. We compared two architectures for classifying the amyloid status and differentiating the three cognitively related phases: Alzheimer’s disease, mild cognitive impairment, and cognitively normal. Our best setup achieved 85.48% (±4.93%) accuracy for predicting amyloid status and 85.42% (±3.44%) accuracy for distinguishing between the three phases. The proposed methods improved upon the best mono-modal setup 73.67% (±4.70%) on T1-weighted MRI and showed competitive performance against a literature reference using FDG-PET. Our results highlight the potential of using non-specific amyloid-related data, like T1-weighted MRI and FDG-PET, for accurate amyloid status prediction.