Parkinson's disease (PD) is a progressive neurodegenerative disorder that affects millions of individuals worldwide, significantly impacting their quality of life. Early and accurate diagnosis of Parkinson's disease is crucial for timely intervention and for personalized treatment plans. Patients with suspected PD undergoes a clinical noninvasive diagnosis, i.e., dopamine transport single-photon emission computed tomography scan (DaTscan) imaging. This chapter focuses on the development and optimization of deep learning models to enhance the accuracy of PD classification based on DaTscan images. The selection of hyperparameters is an important stage in designing the model as it affects the training and performance of deep neural network. Consequently, a whale optimization algorithm (WOA) is used to optimize the hyperparameters in order to train the deep neural network layers. The entire experiment is performed using a publicly accessible NTUA Parkinson dataset, consisting a total of 925 DaTscan images of healthy and PD affected people, with the framework achieving a classification accuracy of 96.27%. To demonstrate the effectiveness of the proposed model, a comparison is drawn of the optimized network with the other networks. This deep learning model can lessen the strain on medical care frameworks and will help in automatic screening of PD patients.