Deep Neural Network Alignment for 3D Tooth Arrangement Generation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
The existing methods for generating a 3D digital model representing a target tooth arrangement in orthodontic treatments are time-consuming, laborious, and skill-dependent, leading to inconsistencies in results.
Innovation Solution
A computer-implemented method using two deep neural networks, where a first network predicts target poses of teeth based on tooth-level feature vectors and a second network improves relative positional relationships using jaw-level feature vectors, with the first network potentially being DGCNN and the second being attention-based, such as Transformer, to generate a digital data set representing a target tooth arrangement.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If manual manipulation of 3D digital model is used to obtain target tooth arrangement, then the operator can adjust the tooth arrangement according to experience, but the process is time-consuming and laborious
Solution Approach 1:
The patent replaces the manual mechanical manipulation of 3D digital models with an automated deep learning system. The first deep neural network automatically predicts target poses of teeth based on initial tooth arrangements, eliminating the need for operators to manually manipulate digital models while maintaining high accuracy through learned patterns from training data.
Solution Approach 2:
The system enables self-service by allowing the deep learning model to automatically generate target tooth arrangements without requiring operator intervention for each case. The model independently processes initial tooth arrangements and produces target arrangements, making the process autonomous and significantly reducing time consumption.
2Measurement precision
If manual manipulation of 3D digital model is used to obtain target tooth arrangement, then the operator can control the transformation process, but the results are strongly dependent on operator's skill and consistency is difficult to ensure
Solution Approach 1:
The patent transforms the subjective, skill-dependent manual manipulation process into an objective, data-driven automated process. By changing from manual control parameters to machine learning model parameters trained on large datasets, the system achieves consistent and reliable results that are not influenced by individual operator skills or experiences.
Solution Approach 2:
The manual control mechanism is replaced with an automated deep learning system that consistently applies learned transformation patterns. This substitution eliminates variability introduced by different operators and ensures reproducible, reliable results across different cases and users.
3Device complexity
If traditional interpolation method is used to generate intermediate tooth arrangements, then the process is simple, but it does not consider occlusal relationships between upper and lower teeth
Solution Approach 1:
The patent merges the processing of upper and lower jaw teeth into a unified deep learning model that simultaneously considers both jaws. The system processes tooth arrangements from both upper and lower jaws together, enabling the model to learn and predict occlusal relationships between corresponding teeth while generating intermediate arrangements, thus combining simplicity with high accuracy.
4Productivity
If deep learning method with two networks is used to generate target tooth arrangement, then the accuracy and efficiency are improved, but the system complexity increases
Solution Approach 1:
The patent segments the complex task of target tooth arrangement generation into two specialized deep neural networks: a first network for predicting individual tooth poses and a second network for refining occlusal relationships. This segmentation allows each network to focus on specific aspects of the problem, improving overall efficiency while managing system complexity through modular architecture.
Data Source
AI summary
The present application provides a computer-implemented method for generating a digital data set representing a target tooth arrangement, comprising: obtaining a first and a second 3D digital models respectively representing upper jaw teeth and lower jaw teeth under an initial tooth arrangement, where the first and the second 3D digital models are in a predetermined relative positional relationship; extracting a tooth level feature vector from each tooth of the first and second 3D digital models; preliminarily aligning the first and second 3D digital models based on the tooth level feature vectors using a trained first deep neural network; extracting a jaw level feature vector for each tooth of the preliminarily aligned first and second 3D digital models; and further aligning the preliminarily aligned first and second 3D digital models to obtain a target tooth arrangement based on the jaw level feature vectors using a trained second deep neural network.