Comparison between 2 Models in Detection of Second Mesio-Buccal Canal of Maxillary First Molars on CBCT Images using Deep Learning

Abstract

Aim: The aim of the present study was to compare two newly developed CNN models to accurately classify MB root canals in maxillary molars to determine presence or absence of MB2 canal on CBCT compared to radiographic assessment performed by expert radiologists.

Methodology: CBCT scans of 41 patients were imported to 3d slicer software to crop the scans with two different methods: the first one is through cropping the MB root only and the second method was through cropping the whole Maxillary molar tooth with its 3 roots (Mesiobuccal – Distobuccal -Palatal). The annotated data was divided into two groups: 80% for training and validation and 20 % for testing. The data was used to develop 2 classification models based on CNN. Confusion matrix and receiver-operating characteristic (ROC) analysis were used in the statistical evaluation of the results of the classification model.

Results: The results of testing the first model (using the images of the cropped Maxillary1st molar with its 3 roots (MB – DB and Palatal) : F1-score,  accuracy, recall and precision values were found to be 0.86, 0.89, 1.0 and 0.75, testing loss was 0.97 and the AUC value was found to be 0.83. While the testing results of using the images of the cropped mesiobuccal root only of Maxillary1st molar : F1-score, accuracy, recall and precision values were found to be 0.93, 0.87, 1.0 and 0.87, testing loss was 0.40 and the AUC value was found to be 0.57.

Conclusion: The performance of training CNN model using images of cropped MB root only or cropped the whole Maxillary molar roots  are both effective in detection of MB2 in maxillary first molar , however, using the latter images showed slightly better performance.