Inferência bayesiana na detecção de potenciais evocados auditivos em regime permanente

Abstract

Auditory steady-state responses (ASSR) are used in clinical practice to assess hearing thresholds. Objective response detection techniques, in the frequency domain, have been developed to identify the ASSR based on the classical Neyman-Pearson approach. These detectors are considered optimal for a given level of significance to either accept or reject the null hypothesis H0 (no response). On the other hand, the Bayesian approach allows the inclusion of prior information for H0 and H1 (response) hypotheses in the model and enables updating of this information with the posterior knowledge. This approach, however, has not been explored with respect to objective ASSR detection techniques. This enables the exploration of new paradigms, which may contribute to this field of study, especially in terms of the time required for response detection. The aim of this work is to investigate the bayesian approach in the development of detectors to better identify the ASSR. Detection algorithms for these potentials were implemented based on the Spectral F test (SFT) and the magnitude squared coherence (MSC), both for the classical and bayesian approaches. Theoretical assessment and Monte Carlo simulations were performed to evaluate the performances of both detectors as a function of the signalto-noise ratio (SNR). To enable the application in ASSR data, a study was carried out on the SNR estimation. Then, the two detectors were applied to ASSR recordings of nine normal-hearing subjects stimulated by amplitude-modulated tones of various intensities. Simulation results showed that the SFT and the MSC performed similarly. Among the scenarios analyzed, the most promising case was the bayesian approach in which the lowest possible values for the a priori probability was selected for the H0, allowing detection at low SNR levels. The bayesian detector worst performance occurred when the a priori probabilities for both hypotheses were equal (reaching ideal performance at SNR levels similar to the Neyman-Pearson detector). Similar results were found in the ASSR data and also showed that higher stimulus intensity led to better performance and faster detection due to improvements in the SNR. It is concluded that the Bayesian detector can be implemented in many ways, given the possibility of arbitrary choices for assigning costs to the decisions that can be made and for the probabilities of occurrence of each of the competing hypotheses. It was found that the strategy of choosing the a priori probabilities has a great influence on the performance that will be achieved by the detector, which in a real application can contribute to reducing the time needed to make a decision.