Published Paper Details:

Multimodal neuroimaging analysis offers promising potential for enhancing the diagnosis of Alzheimer's disease and its subtypes. This work proposes an end-to-end framework that integrates structural MRI, synthetically generated PET images, and cognitive test scores to improve diagnostic accuracy and cognitive phenotype characterization. A 2D CycleGAN model is employed to synthesize PET scans from MRI inputs, preserving structural and metabolic consistency, as validated by SSIM and PSNR metrics. The synthesized PET images are labeled using a fine-tuned Inception-based classifier trained on the Kaggle OASIS dataset, achieving a classification accuracy of 98.08%. Patient-level features from both MRI and PET modalities are extracted using ResNet, while cognitive test scores are used to assign phenotypes through a decision tree classifier. Feature-level fusion is then performed to cluster patients into severity-based groups, highlighting the added value of multimodal integration. This framework demonstrates the potential of multimodal pipelines for Alzheimer's diagnosis, providing a foundation for future clinical decision support systems.