PREDICTING HEARING LOSS PROGRESSION POST-RADIOTHERAPY FOR HEAD AND NECK CANCER USING AUDIOLOGICAL AND DOSIMETRIC MACHINE LEARNING FEATURES

Abstract

Neuroradiation is one of the late effects seen after radiation therapy to the head and neck – it can affect the quality of life and cause functional impairment in the future. In the early identification of patients at high risk of hearing impairment, the role of early intervention in the treatment of patients may be better planned, and proactive measures may be adopted during clinical treatment. This study suggests a machine learning model to predict the progression of hearing loss after radiotherapy based on a set of audiologic and dosimetric parameters. Clinical and treatment-related parameters as well as radiation dose parameters were combined and predictive models were created to determine the patients who will likely suffer from hearing decline following treatment. Multiple machine learning classifiers were evaluated and compared using various classifiers evaluation metrics including Logistic Regression, Random Forest, Support Vector Machine, XGBoosting and Gradient Boosting. The most important predictors were identified by feature importance analysis, and included cochlear mean dose, maximum cochlear dose, baseline hearing threshold, patient age, and treatment duration. Ensemble learning methods outperformed the other methods used in terms of predictive accuracy, as they achieved high accuracy, sensitivity and area under the receiver operating characteristic curve. A dose–response analysis also confirmed that the higher the radiation being delivered to the cochlear, the quicker the hearing loss occurred. The calibration and external validation results showed the robustness and broad applicability of the proposed model across patient populations. The results highlight the need of integrating the results obtained from the audiology assessment with the dosimetric results obtained with radiotherapy in patient risk stratification and patient management. The suggested machine learning approach could help create a clinically applicable decision support tool to minimize the auditory long-term complications and maximize the outcomes for those who have survived head and neck cancers.The answer is found in two areas: Hearing Loss Prediction and Audiological Assessment for Head and Neck Cancer.These are the two areas where the solutions lie: Machine Learning in Hearing Loss Prediction, and Audiological Assessment for Head and Neck Cancer in Radiotherapy Dosimetry.