Deep learning algorithms for detecting fractured instruments in root canals.

Area of Science:

• Dentistry
• Medical Imaging
• Artificial Intelligence

Background:

• Identifying fractured endodontic instruments (FEIs) in periapical radiographs (PAs) is crucial but challenging in root canal treatment (RCT).
• Deep learning (DL) models present a potential solution, but comparative performance analysis is lacking.

Purpose of the Study:

• To evaluate and compare the performance of five deep learning models for detecting FEIs in PAs.
• To identify the most effective DL model for FEI detection in endodontics.

Main Methods:

• A dataset of 700 annotated PAs (381 with FEIs) was used for training, validation, and testing.
• Five DL models (DenseNet201, EfficientNet B0, ResNet-18, VGG-19, MaxVit-T) were trained using transfer learning and data augmentation.
• Performance was assessed using accuracy, AUC, and MCC, with statistical analysis via Friedman test.

Main Results:

• DenseNet201 achieved the highest AUC (0.900) and MCC (0.810) for FEI detection.
• ResNet-18 showed robust results, while EfficientNet B0 and VGG-19 had moderate performance.
• MaxVit-T underperformed significantly; statistical analysis indicated significant differences among models (p < 0.05).

Conclusions:

• DenseNet201 exhibits significant clinical potential for accurate FEI detection.
• ResNet-18 offers a good balance of accuracy and computational efficiency.
• Model-task alignment and optimization are critical for medical imaging applications.