Single-trial Evoked Potential Extraction Toolbox And Algorithm Optimization

The evoked potential(EP)is one of the important means for neurological diseases diagnosis and neuroscience researches.The research of EP is helpful to explore the neural mechanism of physiological and pathological activities.Overlapping averaging filter,currently has become the most widely used method for evoked potentials extraction from ongoing EEG,but this method tends to hide the significant details and information of signals,therefore,resulting in the further development of numerous single-trial evoked potential extraction methods.The key features extracted by the traditional multiple linear regression are homogenized and non-specific.Considering these shortages,an improved multiple linear regression based on the self-adapting feature library was introduced in this study.A MATLAB toolbox was improved in this paper to provide single-trial extraction methods.Meanwhile the structure and human-computer interaction of this toolbox has been optimized.Objective:To remain the dynamics of evoked potentials among trials,improve the time accuracy and preserve more characteristics,a novel single-trial extraction method for evoked potentials was proposed in this study.In order to facilitate the application of the toolbox,the data structure and human-computer interaction should be optimized.Methods:First,wavelet filtering was applied to improve the SNR of evoked potentials,then four groups of single trial EEG data were randomly selected and averaged to perform principal component analysis separately.Then,the principal components were collected to build the feature library.The linear regression analysis,in which independent variables were the components with the highest correlation coefficient with the current data,was performed as the final step to get parameter estimation for single trial extraction.The software structure was simplified based on the MVC design pattern,facilitated the function expansion.The command-line configuration mode was added for the analysis of large amounts of data.The multi-dataset operation function was added to facilitate the data analysis and comparison.Results:The simulated evoked potential extraction results,the latency error is 7.18±2.09 and the amplitude error is 3.90 ± 1.45,shows that all error is lower than that of raw MLRd,ARX,ICA,CSOBI methods.Compared with the benchmark values determined by experts,the new algorithm can achieve more accurate parameter estimation of P300.The difference in latency and amplitude between our results and the benchmark data was 11.16±8.60 ms and 1.40±1.34 ?V.In the real EP data examination,the estimation result of potentials evoked by electrical stimulation on different position were performed t-test,p value were all less than 0.05.Compared with the original,the time consumption was reduced by 25%in multiple datasets process,21%in large amounts of data process,and 44%in wavelet filtering operation.Conclusion:The proposed algorithm with adaptive dynamic feature library can preserve the dynamic characteristics of single trial evoked potentials effectively and improve the accuracy of parameter estimation.Single-trial evoked potential extraction toolbox has increased the operation efficiency effectively.