Detecção do segundo canal mésio-vestibular por meio de redes neurais convolucionais

Abstract

The success of endodontic treatment depends, among other factors, on the location and treatment of all root canals, with the second mesiobuccal canal (MB2) of maxillary molars being one of the main causes of failure when not identified. The aim of this study was to develop and evaluate the performance of a convolutional neural network for the detection of MB2 in maxillary molars on cone-beam computed tomography (CBCT) images. Twenty-two ex vivo maxillary molars were scanned with CBCT, and the volumes were segmented to generate ground truth masks of the MB2 canal. From these volumes, two-dimensional slices were obtained and used to train and test a U-Net model with a DenseNet-121 backbone. Model performance was compared with that of an experienced human observer using accuracy, precision, recall and F1-score as evaluation metrics. The human observer achieved an accuracy of 0.98 and a recall of 0.98 on the test set. The model achieved an accuracy of 0.83, a precision of 0.99 and a recall of 0.66, with reduced performance in more complex cases, especially calcified canals or canals located very close to the main mesiobuccal canal. Under these conditions, the proposed model demonstrated high precision and acceptable overall accuracy for MB2 detection on CBCT slices, suggesting its potential as an auxiliary tool in endodontic practice and image interpretation, although it does not yet replace specialist clinical judgment. . [resumo fornecido pelo autor]