A 3D and Explainable Artificial Intelligence Model for Evaluation of Chronic Otitis Media Based on Temporal Bone...

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Otolaryngology

Background:

• Temporal bone CT scans are crucial for diagnosing chronic otitis media (COM), but interpretation demands expertise.
• Existing AI models for COM lack transparency and may not utilize full diagnostic data from CT scans.

Purpose of the Study:

• To develop an explainable AI system utilizing 3D Convolutional Neural Networks (CNNs) for automated COM evaluation using CT scans.
• To enhance diagnostic accuracy and transparency in COM assessment through advanced AI.

Main Methods:

• Retrospective analysis of temporal bone CT scans from patients with COM.
• Development and training of 3D CNNs for identifying pathological ears and cholesteatoma.
• Performance benchmarking against a 2D model and clinical experts, with validation using heat maps for interpretability.

Main Results:

• The 3D CNN model achieved high accuracy in detecting COM (AUC 0.96) and cholesteatoma (AUC 0.85), outperforming 2D models and expert clinicians.
• The AI system demonstrated strong generalizability across independent datasets and provided interpretable decision-making regions via heat maps.
• Prospective assessment showed the AI system aided preoperative diagnosis with 81.8% accuracy and contributed to clinical decisions in 90.1% of cases.

Conclusions:

• A 3D CNN-based AI model effectively detects COM and cholesteatoma from temporal bone CT scans, surpassing baseline 2D approaches.
• The AI system offers comparable or superior performance to human experts, with enhanced generalizability and interpretability.
• This AI system shows significant potential as a valuable tool for clinicians in real-world COM assessment and diagnosis.