Artificial intelligence for otosclerosis detection on temporal bone CT: a diagnostic study

Abstract

Background: Otosclerosis is a primary osteodystrophy of the otic capsule that causes conductive or mixed hearing loss. High-resolution computed tomography (HRCT) is the preferred imaging tool, but its sensitivity varies and depends on radiologist expertise. Artificial intelligence (AI) may improve diagnostic accuracy by identifying subtle features.

Objective: To evaluate the diagnostic accuracy of a convolutional neural network (CNN) in detecting otosclerosis on temporal bone CT scans.

Methods: This retrospective study included CT scans from 53 surgically confirmed otosclerosis patients and 36 healthy controls. After augmentation, the dataset comprised 74 otosclerosis and 74 control images. A CNN with three convolutional layers, dropout, and a fully connected dense layer was developed. Model performance was assessed using accuracy, sensitivity, specificity, precision, F1-score, and area under the receiver operating characteristic curve (AUC).

Results: The CNN achieved 98% training and 87.5% maximum validation accuracy. In validation, sensitivity was 80.0%, specificity 84.6%, precision 85.7%, F1-score 82.8%, and AUC 0.847. Learning curves showed stable convergence without overfitting.

Conclusion and significance: AI-based CT analysis demonstrated promising diagnostic performance in otosclerosis, especially in cases initially reported as normal. Integration of AI into otologic imaging may enhance diagnostic reliability, surgical planning, and patient management.

Keywords: Otosclerosis; artificial intelligence; convolutional neural network; diagnostic imaging; temporal bone CT.