optimized time-domain feature extraction for early onset diagnosis of parkinson disease from eeg signals

Background: early and accurate diagnosis of parkinson disease (pd) is essential for enhancing patients’ quality of life and enabling more effective symptom management. brain signal analysis, a non-invasive and reliable technique, provides an alternative or complementary method to traditional diagnostic approaches. objectives: this study aims to develop a diagnostic method for pd by combining signal processing techniques with machine learning (ml) algorithms. materials & methods: electroencephalography (eeg) signals were initially segmented into smaller windows using a windowing technique. the intrinsic mode functions (imfs) were subsequently derived using the empirical mode decomposition (emd) technique. the second-order difference plot (sodp) method was applied to each imf, and components with higher informational content were selected for feature extraction. these features were subsequently used to train a decision tree classifier. various window lengths were evaluated to determine the optimal time window for feature extraction, with 4 seconds identified as the optimal duration.results: the proposed method was evaluated using the san diego eeg dataset, which demonstrated state-of-the-art performance compared to existing studies. the classification accuracies achieved for various scenarios were as follows: 99.7% for open-eyes off–pd vs healthy controls (hcs), 96.7% for open-eyes on–pd vs hc, and 98.54% for open-eyes off–pd vs on–pd. conclusion: the results underscore the strong potential of the proposed method in effectively addressing key classification challenges associated with parkinson’s disease.