3D Dento-Maxillofacial Segmentation Using Positional Features
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current methods for 3D classification and modeling of dento-maxillofacial structures using CBCT images face challenges due to unreliable radio density measurements, lack of standardized scaling, and difficulty in distinguishing between teeth and jaw bone structures, leading to inaccurate automatic segmentation.
Innovation Solution
A system and method utilizing deep learning neural networks, specifically 3D convolutional neural networks, that preprocess CBCT data by extracting 3D positional features relative to dental reference planes or objects, enabling accurate classification and segmentation of voxels into jaw, teeth, and nerve tissues without user input.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If manual segmentation methods are used for dento-maxillofacial structures, then the segmentation can be performed with human interpretation, but the process is extremely time-consuming and has low reproducibility
Solution Approach 1:
The patent replaces manual mechanical segmentation processes with automated deep learning neural networks. The system uses 3D CNNs to automatically classify and segment dento-maxillofacial structures from CBCT images, eliminating the need for manual threshold selection and corrections while maintaining high segmentation accuracy and reproducibility.
Solution Approach 2:
The deep learning model performs self-learning and automatic segmentation without requiring continuous human intervention. The neural network is trained on labeled data and then autonomously segments new CBCT scans, making the system self-sufficient for routine segmentation tasks while maintaining consistent results.
2Extent of automation
If deep learning techniques are used for 3D segmentation of dento-maxillofacial structures, then automation is achieved, but accurate segmentation is difficult due to the complexity of the structures and unreliable CBCT density measurements
Solution Approach 1:
The patent transforms the segmentation approach by changing from intensity-based classification to positional feature-based classification. Instead of relying on unreliable Hounsfield Unit density measurements, the system extracts 3D positional features relative to anatomical landmarks and uses these as input features for the neural network, making the segmentation robust against CBCT density variations.
Solution Approach 2:
The patent introduces 3D positional features as an intermediary between the raw CBCT data and the segmentation output. These positional features serve as a reliable mediator that captures anatomical relationships without being affected by the unreliable density measurements inherent in CBCT imaging.
3Use of energy by moving object
If CBCT scans are used for imaging dento-maxillofacial structures, then low-dosage X-ray radiation is applied, but the radio density measurements are unreliable and lack standardized scaling
Solution Approach 1:
The patent extracts and discards the unreliable density information from CBCT scans. By focusing solely on 3D positional features and anatomical relationships rather than Hounsfield Unit values, the system removes the problematic density measurements from the analysis pipeline while maintaining accurate segmentation.
4Device complexity
If traditional 3D CNNs are used for classification, then the network can process 3D data, but it struggles to distinguish between teeth and jaw bone structures due to similar densities
Solution Approach 1:
The patent applies local quality by using different feature representations for different anatomical regions. Instead of treating all voxels uniformly, the system extracts positional features relative to specific anatomical landmarks, allowing the neural network to differentiate between teeth and jaw bone based on their spatial relationships rather than similar density values.
Data Source
AI summary
A computer-implemented method for processing 3D image data of a dento-maxillofacial structure is described wherein the method may comprise the steps of: receiving 3D image data defining a volume of voxels, a voxel being associated with a radiodensity value and a position in the volume and the voxels providing a 3D representation of a dento-maxillofacial structure; using the voxels of the 3D image data to determine one or more 3D positional features for input to a first deep neural network, a 3D positional feature defining information aggregated from the entire received 3D data set; and, the first deep neural network receiving the 3D image data and the one or more positional features at its input and using the one or more 3D positional features to classify at least part of the voxels of the 3D image data into jaw, teeth and/or nerve voxels.