Objectives: Immediately or after head-and-neck (H&N) cancer chemoradiotherapy (CRT), patients may undergone significant sensorineural hearing loss (SNHL) which could affect their quality of life. Radiomic feature analysis is proposed to predict SNHL induced by CRT.
Material and methods: 490 image features of 94 cochlea from 47 patients treated with three dimensional conformal RT (3DCRT) for different H&N cancers were extracted from CT images. Different machine learning (ML) algorithms and also least absolute shrinkage and selection operator (LASSO) penalized logistic regression were implemented on radiomic features for feature selection, classification and prediction. Also, LASSO penalized logistic model was used for outcome modelling.
Results: The predictive power of ten ML methods was more than 70% (in accuracy, precision and area under the curve of receiver operating characteristic curve (AUC)). According to the LASSO penalized logistic modelling, 10 of the 490 radiomic features selected as the associated features with SNHL status. All of the 10 features were statistically associated with SNHL (all of adjusted P-values < .001).
Conclusion: CT radiomic analysis proposed in this study, could help in the prediction of hearing loss induced by chemoradiation. Our study also, demonstrates that combination of radiomic features with clinical and dosimetric variables can model radiotherapy outcome such as SNHL.
Keywords: CT; Chemoradiotherapy; Cochlea; Hearing loss; LASSO; Machine learning; Prediction; Radiomics.
Copyright © 2017 Associazione Italiana di Fisica Medica. Published by Elsevier Ltd. All rights reserved.