Article
Predictive Modeling of Postoperative Performance in Cochlear Implantation: A Machine Learning Approach (en)
* Presenting author
Abstract:
IntroductionHearing loss affects over 17% of the population, significantly impairing quality of life. Cochlear implantation is a major advancement for treating hearing loss, but predicting its effectiveness for individuals is challenging. Developing tools to predict cochlear implant outcomes is crucial.MaterialsWe analyzed data from more than 10500 implantations, including epidemiological factors and preoperative audiometry. The study focused on adults with post-lingual hearing loss, excluding revision implantations, resulting in 2200 cases.MethodsWe developed a decision-tree based machine learning system to predict cochlear implant benefits. The process involved exploratory data analysis, feature selection, data pre-processing, and k-fold cross-validation. The system's interpretability was enhanced through decision tree visualization graphs.ResultsThe system, trained on a large dataset, achieved a mean absolute error of 18.9% on the hold-out dataset, with similar accuracy on recent patient data (2020-2021). Decision tree graphs provided insight into prediction rationale and potential prediction errors.ConclusionThe machine learning system developed for predicting cochlear implant outcomes demonstrated a mean absolute error of 18.9% with a standard deviation of 13.8%. Decision tree graphs offered interpretable prediction paths, aiding clinicians. The results suggest the potential of machine learning in enhancing patient care through better prediction of cochlear implant performance.