An algorithm for automated detection of evoked potentials from polarity reversed electrical stimulations

Abstract

AbstractBackgroundIn neurophysiological research involving neural responses to electrical stimuli, each recording must be searched for evoked potentials (EPs) prior to further analysis. Conducting this process manually is time consuming for the researcher and also inserts bias. However, automated detection methods often struggle to distinguish between artifacts and neural responses, which can have highly complex and varying shapes.MethodsWe have developed a novel algorithm for automated detection of polarity reversed EPs (ADPREP), which uses the knowledge that reversing the polarity of a pair of stimulating contacts reverses the sign of the stimulus artifact, but not that of the neural response. Hence, our algorithm searches for any positive correlation between the recordings from two polarity reversed stimulations, after removing any stimulus decay artifacts. If the peak-to-peak amplitude in a positively correlated region surpasses a user-defined threshold, the recording is labeled as an EP; otherwise, it is not. Neural recordings from deep brain nuclei and cortical regions during deep brain stimulation (DBS) in 28 pediatric patients with dystonia were used to test and prove the validity of the method.ResultsThe ADPREP algorithm is able to distinguish EPs of varying shapes and sizes with a high level of accuracy, as early as 0.35 ms after stimulation, despite large stimulus artifacts. The algorithm has proven useful in initial tests with DBS data in hundreds of stimulation/recording combinations within the basal ganglia and thalamic nuclei as well as from these deep brain nuclei to cortex, at stimulation frequencies up to 250 Hz.ConclusionOur automated EP detection algorithm can accurately detect DBS EPs in deep brain nuclei and cortex, and has promising applications in other stimulation and recording modalities that allow for polarity reversal of the recorded stimulus artifact. The algorithm successfully labels EPs of varying shapes and sizes, as early as a fraction of a millisecond after stimulation, in a range of stimulus frequencies and stimulation-recording pairs – even under large stimulus decay artifacts and in same-lead stimulation and recordings. As such, it is a great method to improve efficacy and minimize human bias, setting up for more reliable conclusions to be drawn from the data.