Predicting Common Audiological Functional Parameters (CAFPAs) as Interpretable Intermediate Representation in a...

Area of Science:

• Audiology
• Machine Learning
• Clinical Decision Support Systems

Background:

• Clinical decision-support systems in audiology can enhance diagnostic objectivity.
• Heterogeneous patient data across local databases presents integration challenges.
• Common Audiological Functional Parameters (CAFPAs) were developed for data integration and interpretability.

Purpose of the Study:

• To automatically derive CAFPAs from patient data using machine learning.
• To predict expert-generated CAFPAs labels with machine learning models.
• To examine the importance of audiological measures for CAFPAs prediction.

Main Methods:

• Prediction of CAFPAs using lasso regression, elastic nets, and random forests.
• Examination and interpretation of audiological measure importance for CAFPAs.
• Evaluation of model generalization to unlabeled data using clustering methods.

Main Results:

• Adequate prediction of ten distinct CAFPAs was achieved.
• All tested machine learning models performed comparably and generalized well.
• Extracted features were plausible, enhancing the interpretability of CAFPAs predictions.

Conclusions:

• Machine learning models can effectively predict CAFPAs, supporting automated audiology diagnostics.
• The developed models facilitate the creation of interpretable clinical decision-support systems.
• This approach enables the extension of audiology clinical decision support to diverse databases.