Analysing and Investigating the Effect of Heroin Use Disorder for the Changes of Distinguishing Features in Brain Signal Processing

Heroin use disorder alters brain function, leading to significant changes in EEG signals. This study proposes a novel approach to identify distinguishing EEG features in heroin addicts and healthy individuals by integrating frequency and non-frequency domain features. The methodology consists of four main stages: (1) Preprocessing, where EEG signals undergo a 50 Hz notch filter and a Butterworth low-pass filter (0.4–45 Hz) to remove noise and artefacts; (2) Feature Extraction, in which both frequency-domain features (power spectral density in alpha, beta, theta, and delta bands) and non-frequency-domain features (approximate entropy, permutation entropy, wavelet entropy, and fractal dimensions of Katz and Petrosian) are computed; (3) Feature Selection, where the Davies–Bouldin index is employed to determine the most discriminative features, independent of the number of clusters; and (4) Analysis and Interpretation, which highlights that approximate entropy in the Cz channel and power spectral density in the upper alpha band in the O1 channel provide the highest discriminative power. Compared to previous studies that primarily rely on frequency-domain features, this approach captures both linear and nonlinear dynamics of EEG signals, leading to improved differentiation between addicted and healthy individuals. While the method demonstrates high accuracy, its sensitivity to EEG preprocessing and the need for larger datasets remain key considerations. The findings suggest that this framework can contribute to more effective addiction diagnosis and monitoring, with potential integration into machine learning-based EEG classification models.