Machine Learning in the Diagnosis and Prognostic Prediction of Dental Caries: A Systematic Review

Abstract

We performed a systematic review to evaluate the success of machine learning algorithms in the diagnosis and prognostic prediction of dental caries. The review protocol was a priori registered in the PROSPERO, CRD42020183447. The search involved electronic bibliographic databases: PubMed/Medline, Scopus, EMBASE, and Web of Science, and grey literature until December 2020. We excluded review articles, case series, case reports, editorials, letters, comments, educational methodologies or assessments of robotic devices and articles with less than 10 participants or specimens. Two independent reviewers selected the studies and performed the assessment of the methodological quality based on standardized scales. We summarize data on the machine learning algorithms used, software, performance outcomes such as accuracy/precision sensitivity/recall, specificity, area under the receiver operating characteristic curve (AUC), and positive/negative predictive values related to dental caries. Meta-analyses were not performed due to methodological differences. Our review included 15 studies (10 diagnostic studies and 5 prognostic prediction studies). Cross-sectional design studies were predominant (12). The most frequently used statistical measure of performance reported in diagnostic studies was AUC value, which ranged from 0.745 to 0.987. For most diagnostic studies, data from contingency tables were not available. Reported sensitivities were higher in low risk of bias prognostic prediction studies (median [IQR] of 0.996 [0.971-1.000] versus unclear/high risk of bias studies 0.189 [0-0.340]; p-value 0.025). While there were no significant differences in the specificity between these subgroups. We concluded that the use of these technologies for the diagnosis and prognostic prediction of dental caries, although promising, is at an early stage. The general applicability of the evidence was limited given that most models were developed outside the real clinical setting with a prevalence of unclear/high risk of bias. Researchers must increase the overall quality of their research protocols by providing a comprehensive report on the methods implemented.

Applications and perspectives of artificial intelligence, machine learning and “dentronics” in dentistry: A literature review

“…This information explosion has become important to develop new mechanics for critical analysis and interpretation of data sets and to achieve effective predictions. [ 9 10 ]…”

“…The majority of these studies allow to perform disease analysis and diagnosis. [ 9 22 23 24 ] Although existing methods can be very robust, ML in dentistry may not replace the dentist but is a method to develop an informed second diagnosis based on a mathematical prediction.…”

“…In this regard, promising results have resulted from the use of these technologies, further encouraging their applicability. [ 5 6 7 8 9 ] For example, some previous studies mentioned that the development of AI seeks to optimize the accuracy of specialists and if possible, surpass it. With the application of AI in the field of implant dentistry, the improvement and performance of dental implants are intended.…”

Early Childhood Predictors for Dental Caries: A Machine Learning Approach

“…Within these classifiers, the decision tree (DT) offers an easy-to-interpret graphical representation, while ensemble algorithms such as random forest (RF) and extreme gradient boosting (XGBoost) are more effective in terms of accuracy and contribute to reducing model overfitting (Kern et al 2019). However, few studies using ML techniques have been conducted to predict dental caries in children and adolescents (Karhade et al 2021; Reyes et al 2022).…”

Development and validation of intraoral periapical radiography-based machine learning model for periodontal defect diagnosis

“…In recent years, innovative computer techniques, especially artificial intelligence (AI), have begun to be used in many areas of dentistry and are helping increase treatment and diagnostic demands. 12–15 The concept of “artificial intelligence” denotes the ability of machines to perform tasks normally completed by humans. 16 Different techniques are used in AI like machine learning and deep learning algorithm.…”