Particle Swarm Optimization of CNN for Early Parkinson’s Disease Detection

Parkinson’sDisease (PD) is a progressive, chronic neurological disease that causes motor dysfunction and for which timely diagnosis is crucial. This study proposes the optimization of Convolutional Neural Networks (CNNs) using Particle Swarm Optimization (PSO) for PD detection based on brain MRI images. CNN models used in the current research trends are typically manually configured, and the process can be time-consuming with the models not being very accurate. This study utilizes PSO, an algorithm that mimics the behavior of swarms, to automate the network architecture optimization for higher diagnosis accuracy. The model was trained on 6166 MRIs of PD patients and healthy controls. PSO also optimally adjusted the hyperparameters during the training phase so that its performance was enhanced. The proposed model gave a classification accuracy of about $98 \%$, a precision of $98 \%$ for the healthy class, and a recall of $97 \%$ for the PD class. It also shows that the application of PSO makes a great improvement to the CNN architecture, and the following PD detection performs better than previous conventional methods. Therefore, it can be stated that the proposed model can be beneficial for early PD diagnosis and provide clinicians with an effective way of analyzing MRI images without the need for human input.