CCZO Residual GhostNet: Parkinson Disease Classification using Optimized Deep Learning Technique

Abstract

Parkinson's disease (PD) classification plays a crucial role in medical diagnosis and patient management. Identifying Parkinson's disease at an early stage can lead to more effective treatment and improved patient outcomes. However, existing methods for Parkinson's disease classification face several limitations. The foremost limitation is the need for accurate and reliable diagnostic tools, as misdiagnosis can lead to inappropriate treatments and unnecessary stress for patients. Thus, a hybrid deep learning model is introduced in this research. The proposed model involves the utilization of EEG signals obtained from a publicly available dataset. Key features are extracted from the EEG signals using a bandpass filter, and every feature is associated with specific brainwave frequencies and cognitive states. The feature mapping and classification are executed through the Chaotic Chebyshev Zebra optimization- based Residual GhostNet (CCZO_Residual_GhostNet). This hybrid classifier, Residual GhostNet, combines ResNet-152 with GhostNet, enhancing classification precision. Furthermore, the CCZO algorithm optimizes the loss function, introducing elements of chaos and Chebyshev mapping to improve classification accuracy. The assessment based on accuracy, sensitivity, specificity, and F-score acquired 98.76%, 98.59%, 98.95%, and 99%, respectively.