| Objectives:The morphological characteristics of food impaction of posterior area with tight proximal contacts were studied based on deep learning,which laid the foundation for follow-up clinical treatment.In this study,by constructing the tooth segmentation network model,the maxillary and mandibular first molars and second molars with high incidence of impaction were accurately segmented,and the relevant parameters were statistically analyzed,and the measured data were analyzed.To explore the causes of tooth impaction and the influence of various characteristic parameters on impaction,so as to provide a basis for clinical treatment of adjacent compact food impaction.Methods:100 patients with food impaction and 100 volunteers without food impaction were recruited to conduct a questionnaire survey.At the same time,the patients and volunteers were made and perfused by doctors who were skilled in clinical operation.Three-dimensional data models of food impacted area and corresponding non-impacted area are obtained by 3-shape intraoral scanner,and point cloud sampling is carried out.On the basis of U-Net,KPConv convolution kernel is combined to replace two-dimensional convolution kernel to construct a network which can efficiently use tooth boundary features to segment teeth.Finally,the local fitting plane is obtained in the sense of least square,and the best segmentation plane is obtained by projecting the segmented point cloud model onto the horizontal and vertical planes,and the related features are solved on this plane.The length of adjacent line,the buccal abduction gap angle,the tongue abduction gap angle,the occlusal abduction gap angle and the adjacent surface area were measured,and the characteristic parameters of impacted and non-food impacted teeth were measured quantitatively,and the database was established.Finally,the upper and lower jaw data were statistically analyzed by normal distribution test,t-test,feature correlation analysis,principal component analysis and binary logistic regression.Results:1.In this experiment,the segmentation network model based on deep learning has a good segmentation effect.The average intersection and union ratio is used as the evaluation index of the segmentation result,and the segmentation accuracy of the three-dimensional tooth model is 92%according to the segmentation network proposed in this paper.2.The adjacent line length,the buccal abduction gap angle,the tongue abduction gap angle,the occlusal abduction gap angle and the adjacent surface area of the maxillary and mandibular with food impaction and without food impaction all met normal distribution.3.There were significant differences in the adjacent line length(P<0.001),the tongue abduction gap angle(P<0.001),the occlusal abduction gap angle(P<0.001)and the adjacent surface area(P<0.001)between maxillary impaction group and non-impaction group.There were significant differences in the adjacent line length(P<0.001),the tongue abduction gap angle(P<0.001),the occlusal abduction gap angle(P<0.001)and the adjacent surface area(P<0.01)between mandibular impaction group and non-impaction group.4.The maxillary impacted group has a 95%reference range for the adjacent line length of(2.917,5.975)mm;a 95%reference range for the buccal abduction gap angle of(47.427,60.735)°;a 95%reference range for the tongue abduction gap angle of(51.277,60.905)°;a 95%reference range for the occlusal abduction gap angle of(51.835,130.811)°;a 95%reference range for the adjacent surface area of(7.484,9.016)mm~2.The maxillary unimpacted group has a 95%reference range for the adjacent line length of(1.914,5.074)mm;a 95%reference range for the buccal abduction gap angle of(43.850,62.384)°;a 95%reference range for the tongue abduction gap angle of(44.780,58.230)°;a 95%reference range for the occlusal abduction gap angle of(9.356,112.632)°;a 95%reference range for the adjacent surface area of(6.848,8.768)mm~2.5.The mandibular impacted group has a 95%reference range for the adjacent line length of(2.271,5.171)mm;a 95%reference range for the buccal abduction gap angle of(53.505,64.626)°;a 95%reference range for the tongue abduction gap angle of(52.014,65.346)°;a 95%reference range for the occlusal abduction gap angle of(62.377,114.600)°;a 95%reference range for the adjacent surface area of(3.319,7.764)mm~2.The mandibular unimpacted group has a 95%reference range for the adjacent line length of(0.640,4.079)mm;a 95%reference range for the buccal abduction gap angle of(52.847,63.614)°;a 95%reference range for the tongue abduction gap angle of(51.586,58.117)°;a 95%reference range for the occlusal abduction gap angle of(30.990,101.376)°;a 95%reference range for the adjacent surface area of(2.964,6.784)mm~2.6.The correlation analysis of maxillary parameters showed that there was a significant positive correlation between the adjacent line length and the tongue abduction gap angle(P<0.01),the occlusal abduction gap angle(P<0.01)and the adjacent surface area(P<0.01).There was a significant positive relationship between the tongue abduction gap angle and the occlusal abduction gap angle(P<0.01).There was a significant positive correlation between the tongue abduction gap angle and the adjacent surface area(P<0.01).There was a significant positive correlation between the occlusal abduction gap angle and the adjacent surface area(P<0.01).7.The correlation analysis of mandibular parameters showed that there was a significant positive correlation between the adjacent line length and the tongue abduction gap angle(P<0.01),the occlusal abduction gap angle(P<0.01)and the adjacent surface area(P<0.01).At the same time,there was a significant positive correlation between the tongue abduction gap angle and the occlusal abduction gap angle(P<0.01).8.Maxillary and mandibular principal component analysis showed that the adjacent line length,the buccal abduction gap angle,the tongue abduction gap angle,the occlusal abduction gap angle and the adjacent surface area were strongly correlated with impaction.9.Maxillary logistics regression analysis showed that the adjacent line length(P<0.01),the tongue abduction gap angle(P<0.01),the occlusal abduction gap angle(P<0.01)and the adjacent surface area(P<0.1)had significant influence on occlusion.Mandibular logistics regression analysis showed that the adjacent line length(P<0.01),the tongue abduction gap angle(P<0.01)and the occlusal abduction gap angle(P<0.05)of the mandible had significant influence on the impaction.Conclusions:1.In this paper,a segmentation network is constructed based on the combination of U-Net network and KPConv convolution kernel proposed by deep learning,and the graph cutting method is used to optimize the segmentation boundary to solve the problem of wrong tooth segmentation and fuzzy segmentation boundary.It basically realizes the automation of tooth segmentation and is suitable for the segmentation of 3D tooth model.2.The adjacent line length,the buccal abduction gap angle,the tongue abduction gap angle,the occlusal abduction gap angle and the adjacent surface area are the main factors affecting the food impaction between maxillary and mandibular first molars and second molars.3.The adjacent line length,the tongue abduction gap angle,the occlusal abduction gap angle and the adjacent surface area had significant positive effects on the food impaction between maxillary first molars and second molars.4.The adjacent line length,the tongue abduction gap angle and the occlusal abduction gap angle have significant positive effects on the food impaction between mandibular first molars and second molars.5.The 95%reference range of The adjacent line length,the buccal abduction gap angle,the tongue abduction gap angle,the occlusal abduction gap angle and the adjacent surface area of upper and lower jaw impaction and non-impaction were counted to provide data reference for later clinical treatment and various types of crown restoration,so as to prevent impaction between artificial crown and natural teeth and improve the accuracy of early diagnosis of impaction. |
PDF Full Text Request