gms | German Medical Science

For improving the diagnosis of hearing impaired patients one recent focus has been on measuring comprehensive test-batteries, i.e. on obtaining quantitative data, such as tone detection thresholds or amplitudes from various objective measures. However, generalizing how a patient with a certain test result is best aided, fitted, or treated remains extremely challenging. Computer models of all kinds have been employed to try to make such a relation but physiologic models require too many parameters to be confined whereas very abstract or black-box models do not offer the causal insights both clinicians and researchers are often looking for. Functional models of the auditory system that often have about 3-10 parameters, such as filter bandwidth or an accuracy limiting internal noise parameter appear to be a good compromise but still require hours of data collection. The question of the present study is therefore how to confine the parameters of a functional model in the most time efficient way. In other words: which experiment needs to be conducted and which experimental conditions need to be measured to learn the most about the model parameters of a patient. For the present experimental proof of concept we tested binaural tone-in-noise detection simply because of being familiar with a very accurate and comprehensive four-parameter model. We propose a likelihood-based fitting procedure, operating in the model-parameter space and providing confidence intervals for the parameters under diagnosis. The procedure is capable of running in parallel to the measurement, and can adaptively set test parameters to the values that are expected to provide the most diagnostic information [1]. Using a pre-defined acceptable confidence interval size, the experiment stops automatically as soon as the size is reached. Three normal-hearing subjects were tested and within 400 alternative-force choice trials (about 25 min) all parameters were determined with relative standard deviations of less than 33%. For comparison: in many cases the estimated parameters differed by more than 33% between subjects, despite similar age and audiograms. The duration to obtain a similar model parameter accuracy with a conventional adaptive stair-case procedure depends on how smart and confined the experimenter chooses the conditions but was multiple times longer in all our cases. The procedure is not limited to psychophysics but can be extended to use and optionally mix all types of experiments.