Deep Learning Accurately Classifies Cochlear Implant Electrode Position on X-Rays

DenseNet121 model classified cochlear implant electrode position on X-rays with an AUC of 0.932 (95% CI 0.836–1.000) and 98.6% specificity, with interpretable attention maps.

• The single-split test set AUC was a mean 0.069 higher than the cross-validation AUC across models, indicating optimistic performance estimates.
• EfficientNet-B3 (weighted) achieved the highest sensitivity at 69.2% (95% CI 42.4–87.3%).
• Grad-CAM confirmed electrode-focused attention in ResNet18 and DenseNet121, supporting interpretability.

Automated summary

RadPigeon summaries are original and for information only. They are not clinical advice.