Title

Portable Abr Acquisition And Automated Classification System Using Pasteless Electrodes

Date of Award

2000

Availability

Article

Degree Name

Doctor of Philosophy (Ph.D.)

First Committee Member

Ozcan Ozdamar, Committee Chair

Abstract

A portable Auditory Brainstem Response (ABR) acquisition and classification system was developed in this study. The portable system consisted of pasteless electrodes, an ultrahigh input impedance bioamplifier and band-pass filter, a data acquisition PCMCIA card, a portable computer and the custom software for ABR acquisition and recognition.The portable battery-powered ABR recording system utilizing pasteless electrodes has several advantages over existing technologies, requiring no skin preparation and paste. Simultaneous recordings utilizing the pasteless electrode/amplifier system and a conventional ABR acquisition system were performed and the results were compared.Two types of neural networks, Multilayer Perceptron (MLP) and Radial Basis Function (RBF), were designed for ABR classification and their performances were compared. MLP trained by the back propagation method was found to have better performance and was finally implemented into the automated classification system.In contrast to most of the previous studies on ABR recognition utilizing neural networks, which use only post-stimulus data, this study used both pre-stimulus and post-stimulus ABR data. New feature extraction methods utilizing peak-to-peak amplitude in running windows and latency were also developed. A rule based module was applied to the output of the neural network and the final classification (Response, No Response and Indeterminate) was given. A noise rejection module using pre-stimulus data was implemented for screening data prior to neural network classification. A total of 21 ears from 16 subjects yielding 1014 recordings were acquired, using the portable system, and were used for the neural network's training and testing. Different neural network models were designed and compared. The final model with the best performance produced 2.8%, 2.2% and 12.8% false positive, false negative and indeterminate recognition rates respectively, and it was implemented into the portable system.

Keywords

Health Sciences, Audiology; Engineering, Biomedical; Artificial Intelligence

Link to Full Text

http://access.library.miami.edu/login?url=http://gateway.proquest.com/openurl?url_ver=Z39.88-2004&rft_val_fmt=info:ofi/fmt:kev:mtx:dissertation&res_dat=xri:pqdiss&rft_dat=xri:pqdiss:9972555