Deep Learning-Based Sinet-V2 Assisted Localization Of Canal Orifices

To establish a preliminary theoretical foundation for real-time clinical assistance in root canal orifice localization by training the SINet-V2 network. Eighty extracted maxillary molars were collected,and microscopic images of the pulp chamber floor were acquired. The root canal orifice regions were manually annotated to construct a standard. Subsequently, photopolymerizable resin was applied over the orifices to simulate calcified deposits on the pulp chamber floor. Microscopic images were reacquired and annotated to generate a simulated situation. Both datasets were divided into training and testing sets at an 8:2 ratio. The standard dataset was used to train the SINet-V2 network for standard orifice localization. The calcified dataset was then employed to train a calcified orifice localization model. Both models were evaluated by comparing their predictions against the gold standard in terms of quantity, morphology, and localization accuracy. The standard orifice model achieved a MIoU of (87.42 ± 5.47)%, MAE of 0.000594 ± 0.000351, NER of 98.44%, and MDis of 1.63 ± 1.40. The calcified orifice model demonstrated an MIoU of (91.87 ± 4.03)%, MAE of 0.000801 ± 0.000543, NER of 96.25%, and MDis of 1.91 ± 1.31. The models exhibited high similarity to the gold standard with minimal localization deviations. The SINet-V2 network based on deep learning effectively localized both standard and simulated calcified root canal orifices in images, providing a preliminary theoretical foundation for AI-assisted real-time root canal orifice localization.