Refining feasibility assessment of endoscopic ear surgery: a radiomics model utilizing machine learning on external auditory canal CT scans.

Abstract

Background: Feasibility assessment of endoscopic ear surgery (EES) relies solely on subjective evaluation by surgeons.

Objective: Extracting radiomic features from preoperative CT images of the external auditory canal, we aim to classify EES patients into easy and difficult groups and improve accuracy in determining surgery feasibility.

Methods: 85 patients' external auditory canal CT scans were collected and 139 radiomic features were extracted using PyRadiomics. The most relevant features were selected and three machine learning algorithms (logistic regression, support vector machine, and random forest) were compared using K-fold cross-validation (k = 5) to predict surgical feasibility.

Results: The best-performing machine learning model, the support vector machine (SVM), was selected to predict the difficulty of EES. The proposed model achieved a high accuracy of 86.5%, and F1 score of 84.6%. The area under the ROC curve was 0.93, indicating good discriminatory power.

Conclusions and significance: The proposed machine learning model provides a reliable and accurate method for classifying patients undergoing otologic surgery based on preoperative imaging data. The model can help clinicians to better prepare for challenging surgical cases and optimize treatment plans for individual patients.