Artificial Intelligence Reading of Postoperative Cochlear Implant X-Rays

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Otolaryngology

Background:

• Postoperative X-rays are standard for confirming cochlear implant (CI) electrode position.
• Current interpretation methods have inter-reader variability and can miss subtle abnormalities.
• Automated classification using deep learning offers a potential solution.

Purpose of the Study:

• To evaluate a deep learning framework for automated classification of CI electrode position on postoperative X-rays.
• To compare the performance of four Convolutional Neural Network (CNN) architectures.
• To assess the potential of these models as clinical support tools.

Main Methods:

• Analysis of 1033 radiographic regions of interest from 673 patients.
• Evaluation of ResNet18, EfficientNet-B0, DenseNet121, and EfficientNet-B3 architectures.
• Utilized patient-level data splitting, transfer learning, consensus ground truth, and five-fold cross-validation.
• Grad-CAM was employed to visualize model attention.

Main Results:

• DenseNet121 (non-weighted) achieved the highest AUC (0.932) and specificity (98.6%).
• EfficientNet-B3 (weighted) demonstrated the highest sensitivity (69.2%).
• Cross-validation showed ResNet18 (weighted) as the most consistent model (AUC 0.861 ± 0.030).
• Grad-CAM confirmed electrode-focused attention in ResNet18 and DenseNet121.

Conclusions:

• Multi-architecture CNNs show potential for classifying CI electrode position with high specificity.
• Clinically interpretable attention maps support further investigation.
• These models show promise as screening-support tools, not autonomous diagnostic systems.