Endodontic Resident University of Nebraska Medical Center College of Dentistry Lincoln, Nebraska, United States
Purpose: Establishing a correct working length is a key prognosticating factor for non-surgical root canal therapy (NSRCT). Learning the skills to perform this step correctly can be a challenge, especially for the novice practitioner. The purpose of this study was to train a deep learning algorithm to assess working length radiographs of single rooted, single canaled teeth in a pre-clinical setting. Materials and
Methods: A search was conducted of extracted teeth receiving non-surgical root canal treatment in the pre-clinical setting at UNMC School of Dentistry. Inclusion criteria are as follows: 1) treatment performed during 1/1/2020 to 1/1/2022; 2) treatment performed in pre-clinical setting; 3) treatment performed by a second-year dental student; and 4) single rooted, single canaled teeth. Exclusion criteria are as follows: 1) working length radiograph not exposed; and 2) working length radiograph of non-diagnostic quality. The radiographs were labeled by the author as short, long, or acceptable. The images were randomized into training and validation sets. A deep learning algorithm was trained using the training set and evaluated using the validation set.
Results: A total of 251 radiographs were included in this study. The trained model correctly labeled the validation set with an accuracy of 70%.
Conclusion: Further research is needed before deep learning can be used as an adjunct to working length radiograph interpretation.
