Automatic threshold detection
thesisposted on 28.03.2022, 18:27 by Jasper Chi Ho Au
The aim of this exploratory project is to decrease the test time for correct detection of auditory responses. A correct detection occurs when a response is correctly identified as a response with a maximum of 5% of non-responses categorised as responses. Determination of auditory thresholds is achieved through quantitative measure of cortical sensory signals but is hindered by the noise present in the recorded signals. This paper concludes that prior information of a subject's waveform response improves denoising for response-present signals, but not for response-absent signals. Gaussian filtering with priors derived from a Gaussian Mixture Model is unable to decrease the number of recordings (test time) for a correct determination of auditory response, when compared with linear averaging and Hotelling's T² statistical test. Weights extracted from Bayesian normalised constants in the filtering process were successfully used to categorise the presence or absence of an auditory response. The artificial bias created by the priors for response-absent signals does not affect the use of these weights as decision criteria.