Multimodal deep learning neuroimaging approach to enhance CT-based diagnosis of Alzheimer's disease.

Neuroimaging plays a critical role in the diagnosis of Alzheimer's disease (AD), with Magnetic Resonance Imaging (MRI) and Positron Emission Tomography (PET) providing detailed structural and functional information for deep learning (DL) based classification. However, their high cost and limited availability restrict widespread clinical use. Computed Tomography (CT), while affordable and widely accessible, is diagnostically insufficient for detecting subtle neurodegenerative changes associated with early AD. To address this limitation, this study proposes a multimodal DL framework that enhances CT-based AD diagnosis by leveraging complementary feature representations learned from MRI. A custom convolutional neural network (CNN) was trained and evaluated using paired CT and MRI data from the Open Access Series of Imaging Studies (OASIS-3). A total of 772 participants with available MRI and CT scans were selected based on Clinical Dementia Rating (CDR) scores and corresponding clinical diagnoses. Participants were categorized as Normal Control (NC) (CDR = 0, n = 300), mild cognitive impairment (MCI) (CDR = 0.5, n = 250), or AD (CDR ≥ 1, n = 222). The overall sex distribution comprised 352 males and 420 females. The CT-only model achieved an accuracy of 84%, with 92% sensitivity and 83% specificity for AD classification. The proposed multimodal model demonstrated superior performance, achieving 92% accuracy, 95% sensitivity, and 91% specificity. Importantly, during CT-only inference, the multimodal framework retained high diagnostic accuracy in identifying disease status, indicating effective transfer of MRI-derived features to CT. These results demonstrate a scalable solution for improving AD detection using CT imaging in resource-limited healthcare.