DeePD-Net: A Deep Learning Approach for Diagnosing Parkinson's Disease Using EEG Signals with IM-CEEMDAN Domain Entropy Features

Parkinson's disease (PD) is a complex, incurable neurodegenerative condition that impacts a significant portion of the global population. Early detection of PD is critically important, as it enables timely and effective interventions that can slow disease progression and improve patients' quality of life. This letter proposes a deep learning-driven approach for early PD diagnosis using EEG (electroencephalogram) signals. In the proposed research, first EEG signals are decomposed into multiple IMFs (Intrinsic Mode Functions) using the proposed IM-CEEMDAN (Improved Complete Ensemble Empirical Mode Decomposition with Adaptive Noise) technique. After that, seven different entropy-based features, namely approximate entropy, sample entropy, bubble entropy, dispersion entropy, slope entropy, permutation entropy, and Rényi permutation entropy, are extracted from the IM-CEEMDAN-decomposed EEG signals to obtain relevant signal features. A deep learning-based model named DeePD-Net is designed and trained with the computed entropy features. The designed model consists of a CNN (Convolutional Neural Network) module, a multi-head attention module, and the proposed NLSTM (Novel Long Short-Term Memory) module. The proposed DeePD-Net model achieves PD detection accuracy of 99.44%. The novelty of the research lies in i) utilization of proposed IM-CEEMDAN for obtaining IMFs from EEG, ii) designing a robust deep learning-based model DeePD-Net for PD detection, iii) integration of multi-head attention mechanism in the DeePD-Net to enhance its PD detection efficacy, and iv) utilization of the proposed robust NLSTM module for PD classification in the designed DeePD-Net model.