Machine learning algorithm based on jaw feature points assist complex maxillary and mandibular reconstruction

Abstract

PurposesLarge-scale jaw reconstruction can hardly achieve satisfactory results only by relying on doctors’ experience. In this study, we assessed a new approach using a machine learning algorithm based on jaw feature points to assist complex jaw reconstruction in patients with maxillary and mandibular defects. MethodsOne hundred and two computed tomography (CT) data on the jaw were collected and 16 skeletal marker points on the jaw were selected. The machine learning algorithm learned the positional relationship between points and built a model, which was used to predict the coordinate position of an unknown point. Then the model was used for a surgical plan in clinical cases. ResultsThe linear regression model based on machine learning can control the error within 3 mm. In linear models, Lasso has a slight advantage over the others. We used Lasso to predict the missing points for two patients with maxillary and mandibular defect, respectively. The operation was carried out as planned, and the defects were successfully repaired. ConclusionsThe restoration of jaw feature points based on a machine learning algorithm is expected to solve large-scale jaw defects without contralateral reference.