Vertical root fracture (VRF) is a prevalent reason for tooth extraction following root canal treatment and even after crown placement. Predicting fractures is challenging due to multifactorial nature. The current study aimed to predict the likelihood of fracture following root canal treatment and crown placement by developing a deep learning (DL) model. DL techniques were employed to analyze a dataset comprising 145 clinical cases consisting of 97 fractured teeth and 48 non-fractured teeth. This dataset spanned a five-year period and encompassed cases involving root canal therapy and crown installation. The analysis identified several root fracture-related parameters, which were incorporated into the DL system. The dataset consisted of 17 features presented in a mixed-type tabular format. The deep neural network (DNN) model surpassed the support vector machine (SVM) model with a higher accuracy (80.7 % vs. 71.7 %) and F1-score value (0.857 vs. 0.817) for predicting root fracture. Furthermore, in determining root fracture occurrence, it was observed that 17 significant characteristics in the DNN model outperformed the 7 features by 11.7 % in accuracy and 10 % in F1-score. DL shows promise in predicting root fracture post root canal therapy and prosthesis, and it may have the potential to aid clinicians in assessing fracture risk and improving decision-making.
Predicting the risk of root fractures following root canal therapy requires diagnosis of the dental history and status of patients. However, dental history is a kind of categorical data type that is not easy to combine with numerical data to obtain good performance in deep learning. The accuracy of support vector machine (SVM) and artificial neural networks (ANNs) is 71.7% and 73.1%, respectively. In this study, a three-stage fusion neural network (TSFNN) is proposed to improve the multiple types of clinical data in the dental field based on ANNs. Clinical data were obtained from 145 teeth, comprising 97 fractured teeth and 48 nonfractured teeth. Each dataset contained 17 items, which were divided into 10 categorical items and 7 numerical items. TSFNN combines numerical and categorical NN with batch normalization and embedding layer techniques and can produce the accuracy of 82.1% and a 19.1% improvement in F1-score. It shows impressive performance in predicting the risk of root fracture. Furthermore, due to the limited amount of clinical data, it is believed that such a pilot study can effectively improve the results when the amount of clinical data is insufficient.