Auditory threshold detection by classifying estimated short latency evoked potentials

Short latency evoked response (SLER) has become a routine clinical tool in neurological and audiological assessment. But, in order to extract SLER from background EEG signal, many repeated single trial measurements are necessary. In some cases, these repetitions are up to 2000. Therefore, measuring period is very time consuming and uncomfortable for subjects. This condition is also limited the SLER usage in clinical applications. In this study, 302 SLER responses extracted by averaging 1024 single trials are used for creating two different data sets. The first set is created from ensemble averaging of 1024 trials for each SLER signals. The second set is obtained from the same single trial measurements by estimating 64 trials of each SLER signal. The support vector machine which is a powerful binary classifier is performed for each data sets for three different feature extraction techniques. In result, the results obtained from estimated data (second data set) classification procedure is better than the results of classical ensemble averaged data set (first data set) with a high accuracy and less time consuming. This results contribute to the SLER usage in clinics more practical than classical ones.