AlexNet CNN Achieves High Accuracy for Detecting Simulated Root Resorption on Radiographs and CBCT

A deep learning model detected simulated external root resorption with 100% accuracy on periapical radiographs and up to 97.92% on CBCT. Performance dropped to 73.96% with three-dimensional reconstruction, especially in treated teeth.

• In vitro study using 480 bovine incisors divided into groups with/without root canal treatment and simulated resorption.
• CNN based on pre-trained AlexNet and adapted with transfer learning.
• CBCT with blooming artifact reduction (BAR) improved accuracy (97.92% vs 95.83% without BAR).

Automated summary

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