MACHINE LEARNING-ASSISTED DIAGNOSIS OF TEMPORAL BONE AND EAR CANAL MALIGNANCIES USING CT IMAGING AND CLINICAL SYMPTOMS

Abstract

The aim of this study was to present a machine learning-based diagnostic system, which combines computed tomography (CT) imaging features and clinical symptom data to improve the diagnosis and classification of temporal bone and ear canal malignancy. This study was a retrospective cohort study of patients that presented with lesions of the temporal bone and external auditory canal. Bone erosion patterns, soft tissue extension, margin of the lesion, involvement of the middle ear, mastoid infiltration, and clinical variables which included otalgia, otorrhea, hearing loss, facial nerve dysfunction, tinnitus and duration of symptoms were extracted from the CT images. Stratified cross validation was used to train and test the machine learning models with various machine learning algorithms including Random Forest, Support Vector Machine, Gradient Boosting and Extreme Gradient Boosting (XGBoost). The diagnostic performance was assessed by accuracy, sensitivity and specificity, precision, F1 score and Area under the receiver operating characteristic curve (AUC). The performance of the proposed framework was acceptable with best performance given by the ensemble-based models as compared to the conventional models. XGBoost achieved the best classification performance and provided an accurate classification of the temporal bone malignant and benign lesions.