MRI AND CT RADIOMICS FOR PREDICTING CHOLESTEATOMA RECURRENCE AND SURGICAL COMPLEXITY IN CHRONIC EAR DISEASE

Abstract

Cholesteatoma is a chronic destructive ear disease, a direct consequence of which is bone erosion, frequent postoperative recurrence, hearing loss, and the risk of hearing loss. The risk of recurrence and the complexity of the surgery are difficult to predict accurately preoperatively, due to the fact that visual interpretation of conventional imaging techniques may be insufficient to accurately represent subtle patterns of disease, tissue behavior, and anatomical involvement. The aim of this study is to develop an MRI- and CT-based radiomics framework for predicting the recurrence of cholesteatoma in patients with chronic ear disease and predicting the complexity of surgery. The quantitative radiomic features associated with lesion morphology, texture heterogeneity, bone erosion, soft-tissue extension, mastoid involvement, and middle ear structural changes were extracted from the computed tomography and magnetic resonance imaging scans obtained preoperatively. Clinical and imaging-derived variables were combined with machine learning models to categorize patients based on risk of recurrence and anticipated surgical complexity. Clinical and imaging-derived variables were used to create machine learning models to categorize patients by risk of recurrence and expected surgical difficulty. Logistic Regression, Random Forest, Support Vector Machine, Gradient Boosting and XGBoost algorithms were trained and evaluated using various metrics such as accuracy, precision, recall, F1 score, and area under the receiver operating characteristic curve. The proposed radiomics-based models exhibited high predictive performance, and the ensemble classifiers exhibited the best discriminatory ability among the traditional classifiers. The lesion texture heterogeneity, erosion of the ossicles, destruction of the scutum, opacification of the mastoid, diffusion restriction patterns, and middle ear involvement were among the most important predictors seen using feature importance analysis. The results indicate that multimodal radiomics can offer objective, reproducible, and clinically useful preoperative risk stratification biomarkers. This technique can help in treatment planning, post-surgical monitoring for recurrence, patient counseling, and surgical planning for chronic ear disease in otologic surgery.