Multimodal SVM Classification for Early-stage Alzheimer's Disease Diagnosis Using T1-weighted MR and F-18 FDG PET Imaging

Alzheimer's disease is a significant global health challenge characterized by progressive brain degeneration associated with aging. Early detection is crucial for improved prognosis and treatment outcomes. Our study aimed to develop a support vector machine (SVM) classification model using T1-weighted MR and F-18 FDG PET brain imaging to classify cognitive normal (CN) and early-stage Alzheimer's disease, including early mild cognitive impairment (EMCI) and late mild cognitive impairment (LMCI), in individuals aged 65-75. The study comprised three steps. Firstly, image preprocessing involved uniformity processing, B1 bias field correction for MR images, and FWHM optimization with spatial/intensity normalization for PET images. Secondly, MR images were registered to the MNI305 brain template for structural segmentation, and PET images were normalized using the standardized PET template. Co-registration of normalized PET images to segmented MR images provided anatomical segmented PET uptake volumes. These served as input for the classification model. Thirdly, SVM models classified CN vs. LMCI, EMCI vs. LMCI, and CN vs. EMCI using MRI, PET, and combined PET/MR. Feature sets included all features, clinical-based features, and F1-score ranked features, resulting in 27 classification models. In CN vs. EMCI, combined PET/MR with all features achieved 0.71 AUC, 65.78% accuracy, and 68.67% specificity. For CN vs. LMCI, superior performance was observed with combined PET/MR using F1-score ranked features: 0.82 AUC, 77.78% accuracy, and 77.55% specificity. In EMCI vs. LMCI, PET alone with all features achieved the highest performance: 0.72 AUC, 69.23% accuracy, and 60.32% specificity. In conclusion, PET is pivotal in MCI stage differentiation, and using all features aids in challenging tasks (CN vs. EMCI and EMCI vs. LMCI). The combined PET/MR modality notably distinguishes CN from MCI, emphasizing the potential of multimodal imaging to enhance differentiating cognitive states in individuals.