Interpretable AI Explores Effective Components of CAD/CAM Resin Composites

Abstract

High flexural strength of computer-aided manufacturing resin composite blocks (CAD/CAM RCBs) are required in clinical scenarios. However, the conventional in vitro approach of modifying materials’ composition by trial and error was not efficient to explore the effective components that contribute to the flexural strength. Machine learning (ML) is a powerful tool to achieve the above goals. Therefore, the aim of this study was to develop ML models to predict the flexural strength of CAD/CAM RCBs and explore the components that affect flexural strength as the first step. The composition of 12 commercially available products and flexural strength were collected from the manufacturers and literature. The initial data consisted of 16 attributes and 12 samples. Considering that the input data for each sample were recognized as a multidimensional vector, a fluctuation range of 0.1 was proposed for each vector and the number of samples was augmented to 120. Regression algorithms—that is, random forest (RF), extra trees, gradient boosting decision tree, light gradient boosting machine, and extreme gradient boosting—were used to develop 5 ML models to predict flexural strength. An exhaustive search and feature importance analysis were conducted to analyze the effective components that affected flexural strength. The R2 values for each model were 0.947, 0.997, 0.998, 0.983, and 0.927, respectively. The relative errors of all the algorithms were within 15%. Among the high predicted flexural strength group in the exhaustive search, urethane dimethacrylate was contained in all compositions. Filler content and triethylene glycol dimethacrylate were the top 2 features predicted by all models in the feature importance analysis. ZrSiO4 was the third important feature for all models, except the RF model. The ML models established in this study successfully predicted the flexural strength of CAD/CAM RCBs and identified the effective components that affected flexural strength based on the available data set.

Machine Learning Analysis of Microtensile Bond Strength of Dental Adhesives

“…The size of a data set can significantly affect the performance of ML models. Compared to many other fields, dental material data sets are typically small and complex (Li et al 2022). Small data sets could result in overfitting, which happens when a model fits too well to the training data but does not generalize well to new data (Ying 2019).…”

“…The features should uniquely define various materials regarding the target property (Ward et al 2016). Chemical compositions have been commonly used as input features to predict material properties using ML (Hu et al 2020; Li et al 2022).…”

Influence of print orientation and wet-dry storage time on the intaglio accuracy of additively manufactured occlusal devices

“…43,44 Artificial intelligence (AI) models are being developed for optimizing the manufacturing procedure, which may assist in establishing the optimal printing protocol based on the manufacturing trinomial and clinical application of the device being printed. 45,46 Limitations of the present study included the limited vat-polymerization technologies, printers, and materials tested, as well as the limited print orientations assessed. Additionally, a novel storage protocol was used.…”

Materials informatics for developing new restorative dental materials: a narrative review

“…In the dental field, Li et al were the first to apply the MI approach to predict the flexural strength of computer-aided design/computer-aided manufacturing (CAD/CAM) resin composites, and they successfully explored the optimum compositions to achieve desirable flexural strength (19). Thus, the MI approach promises to make dental material research more efficient than the conventional trialand-error approach (20).…”