Related Experiment Videos
Neural networks applied to retrocochlear diagnosis
D E Callan1, R E Lasky, C G Fowler
1University of Wisconsin-Madison, Department of Communicative Disorders, USA. dcallan@hip.atr.co.jp
Summary
Neural networks improve audiological test batteries for retrocochlear pathology diagnosis. Combining auditory brainstem evoked response (ABR) with other tests enhances accuracy beyond individual test performance.
Area of Science:
- Audiology
- Diagnostic performance evaluation
- Signal detection theory
Background:
- Traditional test batteries may underperform compared to individual diagnostic tests.
- Previous studies often weighted all tests equally, limiting diagnostic accuracy.
- Signal detection theory provides a framework for evaluating test performance.
Purpose of the Study:
- To apply neural networks for differentially weighting audiological tests in a battery.
- To evaluate the diagnostic performance of audiological tests for predicting retrocochlear pathology.
- To improve upon previous test battery methodologies.
Main Methods:
- Utilized neural networks to analyze audiological test data.
- Applied differential weighting to individual test results within a battery.
- Compared the diagnostic accuracy of various test combinations and raw measures.
Main Results:
- The auditory brainstem evoked response (ABR) was confirmed as a superior individual test for retrocochlear disease.
- Combining ABR with other audiological tests (acoustic reflex, tone decay, word recognition) improved identification accuracy.
- Using raw test measures with ABR outperformed dichotomous (positive/negative) diagnostic approaches.
Conclusions:
- Neural networks offer a superior method for optimizing audiological test batteries.
- Combining ABR with other audiological tests enhances diagnostic accuracy for retrocochlear pathology.
- Raw test measures integrated with ABR provide more precise diagnostic information.