Autism EEG Signal Pre-Processing: Performance Evaluation of MS-ICA and Butterworth Filter

Autism Spectrum Disorder (ASD) is a neurological condition characterized by challenges in communication and social interaction, accompanied by the development of repetitive behavioral patterns. Electroencephalography (EEG) is primarily used to assess brain function in children with Autism Spectrum Disorder (ASD), mainly due to its non-invasive nature and superior temporal resolution compared to other neuroimaging methods. However, EEG signals are often contaminated by biological artifacts, such as eye movements and muscle contractions, which can significantly distort analysis outcomes. Pre-processing is therefore required to increase the accuracy of the EEG signal before additional analysis. The goal of this study was to compare and evaluate the performance of two pre-processing techniques, the Butterworth Band-Pass Filter and Multiscale Independent Component Analysis (MS-ICA), using four different performance metrics: Mean Absolute Error (MAE), Mean Squared Error (MSE), Root Mean Squared Error (RMSE), and Signal-to-Noise Ratio (SNR). The Butterworth method has an MAE of 227.57, which is acceptable. However, it produced an MSE of 160,653.22, an RMSE of 394.49, and a maximum SNR of only 1.33 dB. MS-ICA performs far better with a best MAE of only 0.44, an MSE of 3.33, an RMSE of 1.76, and an SNR of 30.88 dB. Paired t-test (p < 0.05) was employed to determine statistical significance,  while Cohen's d was used to assess the practical significance of the results. The effect sizes of MAE (d = 1.60), MSE (d = 1.02), RMSE (d = 1.54), and SNR (d = -9.50) were all calculated as large. These findings demonstrate that MS-ICA offers both statistical advantages and strong practical usefulness for noise removal while preserving the structural integrity of the original EEG signals. Therefore, MS-ICA proves to be the best approach for pre-processing EEG signals to be used for analysis in children with ASD