3D Dental Object Canonical Pose Alignment Using Deep Learning
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing 3D image registration techniques for dental structures, particularly using CBCT and IOS data, face challenges in accurately aligning and superimposing data sets due to variance in data formats, modalities, and coordinate systems, requiring human intervention and being computationally expensive.
Innovation Solution
A 3D deep neural network is trained to determine a canonical pose for 3D dental structures, transforming different data sets into a standardized coordinate system for automated superimposition, using convolutional neural networks to process voxelized data and align multiple data sets without human interaction.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If traditional 3D image registration techniques are used for aligning CBCT and IOS data sets, then human intervention is required and computational cost is high, but accuracy and reliability of superimposition are limited
Solution Approach 1:
The patent replaces traditional mechanical/image-processing-based registration methods with a deep learning neural network system. The neural network automatically learns to align CBCT and IOS data sets by training on labeled examples, eliminating the need for manual intervention while achieving high registration accuracy through automated feature extraction and transformation parameter optimization.
Solution Approach 2:
The system performs self-service by automatically aligning data sets without requiring user input or manual adjustment. The neural network independently processes input data, determines optimal transformation parameters, and outputs aligned results, making the registration process autonomous and efficient.
2Productivity
If traditional 3D image registration techniques are used, then human intervention is required, but computational complexity and processing time increase
Solution Approach 1:
The patent substitutes computationally intensive traditional image registration algorithms with a neural network-based approach. The neural network, once trained, performs registration through efficient forward propagation and parameter optimization, significantly reducing processing time and computational resource requirements compared to iterative traditional methods.
3Object-affected harmful factors
If CBCT data is used for 3D modeling, then radiation dose and acquisition cost are reduced, but contrast and differentiation between structures are poor
Solution Approach 1:
The patent merges CBCT volumetric data with IOS surface scan data to create a comprehensive 3D model. The neural network integrates information from both modalities, combining the radiation-free surface detail of IOS with the volumetric bone structure from CBCT, thereby overcoming the contrast limitations of low-dose CBCT while maintaining low radiation exposure.
Solution Approach 2:
The deep learning neural network acts as an intermediary that processes and fuses data from different modalities. It learns to complement the strengths of each data type and mitigate their weaknesses, transforming low-contrast CBCT data into high-precision 3D models by integrating IOS surface information.
4Object-affected harmful factors
If optical scan data is used instead of CBCT, then radiation is eliminated and spatial resolution is improved, but differentiation between teeth and gingival regions is not achieved
Solution Approach 1:
The patent combines optical scan data with CBCT data to create a comprehensive 3D model that includes both surface detail and internal structure information. The neural network integrates the radiation-free high-resolution surface data from optical scanning with the volumetric tissue differentiation data from CBCT, achieving both low radiation exposure and complete tissue differentiation.
Data Source
AI summary
A method for automatically determining a canonical pose of a 3D object comprises: providing one or more blocks of voxels of a voxel representation of the 3D object to a first 3D deep neural network, the first 3D neural network being trained to generate canonical pose information; receiving canonical pose information from the first 3D deep neural network, the canonical pose information comprising for each voxel a prediction of a position of the voxel in the canonical coordinate system; using the canonical coordinates to determine an orientation and scale of the axes of the canonical coordinate system and a position of the origin of the canonical coordinate system relative to the axis and the origin of the first 3D coordinate system and using the orientation and the position to determine transformation parameters of the first coordinate system into canonical coordinates; and, determining a canonical representation of the 3D dental structure.