Modelo de inteligência artificial para análise do prognóstico do crescimento maxilofacial de pacientes com fissura labiopalatina: desenvolvimento e validação

Abstract

Cleft lip and palate are the most common congenital deformities affecting the face; their rehabilitative treatment is lengthy, costly, complex, and requires specialized professionals. Analyzing the maxillofacial growth prognosis of cleft patients is crucial to guide the orthodontic treatment plan and should be done by an orthodontist with expertise in orthodontic correction of cleft patients. In this context, adopting new technologies in the health sector aims to make processes more efficient, break geographical barriers, contribute to professional training, and improve clinical decision-making. This study aimed to develop and validate a deep learning model for analyzing the maxillofacial growth prognosis of individuals born with cleft lip and palate, based on the GOSLON occlusal indices. Accordingly, this study constructed an intelligent model using transfer learning in a Convolutional Neural Network (CNN) trained with 589 images of intraoral scans of patients with unilateral cleft lip and palate, classified according to the GOSLON Yardstick index by 5 previously calibrated experienced orthodontists. Through the phases of preprocessing, training, testing, and validation, eight different CNNs that had a record in the literature of performing medical image classification tasks were tested. The best-performing model was VGG16, with a precision of 0.904, accuracy of 0.90, and an F1-score of 0.89. After validation, a new concordance test was conducted comparing the model's results with the analysis of two orthodontists, obtaining a kappa of 0,983. The implementation of technologies in health, as well as in the field of cleft lip and palate, depends on the tool's performance, necessity, and integration into the workflow. VGG16 demonstrated the capability to classify intraoral arches following the GOSLON Yardstick index accurately and presents lower computational cost, thus being applicable in all uses of the indices such as: supporting decision-making regarding the orthodontic treatment of patients with unilateral cleft lip and palate, as a metric for evaluating the quality of clinical protocols, and as a tool for estimating treatment costs.