Page 1

birpublications.org/dmfrDentomaxillofacial Radiology (2023) 0, 20220209© 2023 The Authors. Published by the British Institute of RadiologyRESEARCH ARTICLEDetection of the separated root canal instrument on panoramicradiograph: a comparison of LSTM and CNN deeplearning methods1Cansu Buyuk, 2Burcin Arican Alpay and 3Fusun Er1Department of Dentomaxillofacial Radiology, Istanbul Okan University, Faculty of Dentistry,İstanbul, Turkey; 2Department ofEndodontics, Bahcesehir University, Faculty of Dentistry,İstanbul, Turkey; 3Information Systems Engineering, Piri Reis University,Faculty of Engineering,İstanbul, TurkeyObjectives:A separated endodontic instrument is one of the challenging complications of rootcanal treatment. The purpose of this study was to compare two deep learning methods that areconvolutional neural network (CNN) and long short-term memory (LSTM) to detect the separatedendodontic instruments on dental radiographs.Methods:Panoramic radiographs from the hospital archive were retrospectively evaluated by twodentists. A total of 915 teeth, of which 417 are labeled as “separated instrument” and 498 are labeledas “healthy root canal treatment” were included. A total of six deep learning models, four of whichare some varieties of CNN (Raw-CNN, Augmented-CNN, Gabor filtered-CNN, Gabor-filtered-augmented-CNN) and two of which are some varieties of LSTM model (Raw-LSTM, Augmented-LSTM) were trained based on several feature extraction methods with an applied or not applied anaugmentation procedure. The diagnostic performances of the models were compared in terms ofaccuracy, sensitivity, specificity, positive- and negative-predictive value using 10-fold cross-validation.A McNemar’s tests was employed to figure out if there is a statistically significant difference betweenperformances of the models. Receiver operating characteristic (ROC) curves were developed toassess the quality of the performance of the most promising model (Gabor filtered-CNN model) byexploring different cut-off levels in the last decision layer of the model.Results:The Gabor filtered-CNN model showed the highest accuracy (84.37 ± 2.79), sensitivity(81.26 ± 4.79), positive-predictive value (84.16 ± 3.35) and negative-predictive value (84.62 ± 4.56with a confidence interval of 80.6 ± 0.0076. McNemar’s tests yielded that the performance of theGabor filtered-CNN model significantly different from both LSTM models (p< 0.01).Conclusions:Both CNN and LSTM models were achieved a high predictive performance onto distinguish separated endodontic instruments in radiographs. The Gabor filtered-CNN modelwithout data augmentation gave the best predictive performance.Dentomaxillofacial Radiology(2023)0, 20220209. doi: 10.1259/dmfr.20220209Cite this article as:Buyuk C, Arican Alpay B, Er F. Detection of the separated root canalinstrument on panoramic radiograph: a comparison of LSTM and CNN deep learningmethods.Dentomaxillofac Radiol(2023) 10.1259/dmfr.20220209.Keywords:Artificial intelligence; CNN; deep learning; LSTM; panoramic radiograph; sepa-rated endodontic instrumentsIntroductionThe common causes offailure in endodontic treat-ments are the persistence of bacteria, inadequate rootcanal filling, poor obturation quality, microleakage, andcomplications ofinstrumentation such as separatedendodontic instruments (SEI).1 The incidence of theSEI in the root canal was reported to range from 0.4 to7.4%.2 In teeth with a SEI, the success of retreatmentCorrespondence to: Cansu Buyuk, E-mail: cansubuyuk@yahoo.comReceived 14 June 2022; revised 08 December 2022; accepted 05 January 2023;published online 24 January 2023