Modified graph neural network-oriented optimization model for the classification of PD.

Methods for categorizing Parkinson's Disease (PD) aim to automatically distinguish among healthy individuals as well as those who have the condition in addition to identifying the illness's phases or subtypes. These methods enable early detection as well as customized treatment using information from wearable sensors, MRI, DaTscan, and clinical assessments. However, there are still challenges due to the fact that symptoms from many conditions are identical, disease progression is inconsistent, there are few labelled datasets, as well as multi-modal information is diverse. Furthermore, it is sometimes difficult to recognize early-stage mild indications. To provide accurate as well as therapeutically meaningful findings across a range of patient groups, developing robust classification methods requires effective feature extraction, data integration, as well as generalization procedures. Therefore, this paper proposes the Classification of Parkinson's Disease (CPD) model with the help of deep learning methodology. The dataset is initially collected from standard reputed sources called Parkinson's Disease Classification Benchmark (PDCB) dataset. This collected data undergoes pre-processing by the multimodal data alignment and normalization approaches. From these pre-processed data, the segmentation is done by the Shifted Window UNETR (Swin UNETR) approach. Next, the features are extracted from these segmented data using the Short-Time Fourier Transform (STFT) method. The extracted features enter the final classification process using novel Modified Graph Neural Network (MGNN). The parameter tuning in traditional GNN is performed by nature inspired optimization algorithm known as Fossa Optimization Algorithm (FOA). Accuracy maximization is considered as the objective function for this considered CPD model. This proposed MGNN-FOA of the CPD model classifies the final output into four classes such as Healthy Control, Early-Stage, Mid-Stage and Late-Stage respectively. The proposed MGNN-FOA is 12.52% and 12.68% superior to the other considered conventional models with respect to accuracy and recall respectively for the CPD model.