Prediction of failures and indication for replacement of restorations in primary and permanent teeth through machine learning

• Grant number: 22/16528-3
• Support Opportunities: Scholarships abroad - Research
• Start date: July 05, 2023
• End date: July 01, 2024
• Field of knowledge: Health Sciences - Dentistry
• Principal Investigator: Fausto Medeiros Mendes
• Grantee: Fausto Medeiros Mendes
• Host Investigator: Marie Charlotte Dymphna Nicole Joseph Martine Huysmans
• Host Institution: Faculdade de Odontologia (FO). Universidade de São Paulo (USP). São Paulo , SP, Brazil
• Institution abroad: Radboud University Medical Center (Radboudumc), Netherlands

Abstract

Criteria used for indicating replacement or predicting restoration failures are subjective and based on weak evidence, usually leading to unnecessary replacement. However, with the advent of artificial intelligence, using prediction models created by machine learning (ML) techniques could minimize this problem, aiding and driving clinicians to more conservative management procedures. Therefore, the study will aim to develop, validate, and test ML algorithms to predict indication of replacement and failures of restorations in primary and permanent teeth. The specific aims will be: (i) to develop ML algorithms to predict the indication of restorations replacement made by dentists from The Netherlands; (ii) to develop and train ML models for diagnosis and indication of restorations interventions based on bitewings through crowdsourced annotations made by dentists from Brazil and the Netherlands; (iii) to develop and validate ML algorithms to predict restorations failures in primary and permanent teeth; (iv) to test the developed algorithms in an independent retrospective sample of patients treated by dentists from the Netherlands. For aims (i), (ii), and (iii), we will use retrospective data from two randomized clinical trials conducted in Brazil, and a practice-based study conducted with Dutch dentists. Artificial intelligence approaches based on supervised ML (aims I and iii) and on convolutional neural networks (aim ii) will be employed. For aim (iv), accuracy parameters obtained with the application of these algorithms in this independent retrospective sample will be evaluated. (AU)