Enhancing Parkinson's Disease Detection from Brain MRI through Optimized Super Resolution Generative Network

Introduction: Parkinson's disease (PD), a neurodegenerative disorder that results in motor symptoms like stiffness and tremors, is characterized by the death of dopaminergic neurons in the substantia nigra of the brain. Early and accurate diagnosis of PD is challenging due to overlapping symptoms with other neurological conditions and the limitations of traditional diagnostic methods. Objectives: This study proposes a novel approach utilizing advanced deep learning techniques to enhance PD diagnosis. Methods: A Super-Resolution Generative Adversarial Network (SRGAN) was developed, integrating a customized DeepResNet (ResNet-150 with skip connections), Generative Adversarial Networks (GANs), and Gaussian filtering to improve the quality of MRI and PET imaging data. Image preprocessing steps including contrast enhancement, noise reduction, gamma correction, and pixel value normalization were employed to ensure high-quality input for the model. Results: The SRGAN demonstrated superior performance compared to existing convolutional neural network (CNN) models such as DenseNet, AlexNet, VGG 19, and ResNet, achieving an accuracy of 96.5%, an F1 score of 94.5%, and a balanced accuracy (BAAC) of 93.5%. Additionally, the Skip Connection Optimization (SCO) within the SRGAN outperformed other optimization techniques like Genetic Algorithms (GA), Ant Colony Optimization (ACO), Grey Wolf Optimizer (GWO), Stochastic Gradient Descent (SGD), ADAM, and AdaMax. Conclusions: The results highlight the effectiveness of advanced deep learning models in improving the diagnosis and potential treatment strategies for Parkinson's disease.