Enhancing ERP component detection by estimating ERP latency variability using hidden process model

In this paper, we propose an approach to improve detection of event related potential (ERP) component using hidden process model, which enables estimating the trial-to-trial variability of ERP latency to overcome limitation of the conventional averaging method for extracting ERP components. By using HPM, which is a generative model for estimating underlying process that has unknown onset timing, we can estimate responses of assumed processes underlying cognitive functions and the probability distribution of the onset timing of each process. We applied HPM to ERP data obtained during the oddball task and distinguished ERPs induced by target or nontarget stimuli. We designed 2-process and 3-process HPMs and estimated the responses of each process in these HPMs. Then, we compared these responses with ERP waveforms obtained by conventional averaging. As a result, the waveforms of the estimated response from each model resembled that of averaged ERP while the peak amplitude was higher in estimated responses than in averaged ERP. In addition, the difference of the area under curve between target and nontarget condition was clearer in estimated responses than in averaged ERP. This suggests that HPM might be able to overcome the latency variability of ERP components to estimate more exact components, which will enhance differentiating ERP components between conditions in an ERP study.