Efficient Multi-Modal Image Fusion Deep Learning Techniques for Classifying Neurodegenerative Disease Type

Neurodegenerative disorders like Parkinson's and Alzheimer's represent a notable menace to the welfare of humanity. They primarily result from the progressive deterioration of the peripheral and central nervous systems, significantly affecting an individual's daily life. To diagnose these disorders, ongoing clinical assessments are necessary. Modern medical diagnosis often employs deep learning techniques. One challenge with the deep learning approach is handling diverse datasets of multiple modalities. Earlier research relied on just one modality, making it an inadequate diagnostic aid. In this work, we combined the benefits of various modalities to create a hybrid technology designed for practice in the timely identification of Parkinson's disease. This work employs an MRI image dataset and a DaT scan dataset connected to Parkinson's disease. This study develops four alternative models. The general framework of the several recommended techniques is as follows: first, we use picture augmentation by methods such as blurring and sharpening. Subsequently, we either engage in early image fusion, transferring the fused images for feature extraction and subsequent classification, or independently extract features from both modalities and later fuse these independently extracted features before conducting the classification process. Finally, we performed a comparative analysis between the state-of-the-art model chosen as the baseline and the several models that were put forth. Among these, Model 2 exhibited superior performance, achieving a test accuracy of 93.96%.