3D Root Shape Prediction Using Deep Learning

Resolve Bottlenecks,
Find Innovative SolutionsGenerate Solutions

Solution Overview

Problem

Current methods for predicting 3D root shapes in dental applications are limited by the accuracy and reliability of existing systems, which often require specific 3D reference models for specific tooth and patient classes, and are challenged by the variability and inconsistency of CBCT data, leading to inaccuracies in anatomical representation.

Innovation Solution

A computer-implemented method using a 3D deep neural network trained on clinical data to predict anatomically accurate 3D root shapes from 3D crown data, without the need for specific tooth or patient class inputs, by transforming and processing voxel representations to generate a complete tooth model.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Ease of operation

If crown morphology-based mathematical reference models are used to approximate root shape, then the system can provide a 3D root model without X-ray data, but the accuracy and reliability are limited with relatively large deviations from the actual root shape

Engineering Contradiction:

Improveability to generate root model without X-ray dataVSAvoidaccuracy of root shape approximation

Core Design Contradiction:

Ease of operationVSMeasurement precision

Solution Approach 1:

The patent transforms the input crown geometry into a standardized voxel representation with specific dimensional parameters (e.g., 64x64x64 voxels), then uses these standardized parameters as input to the neural network. This parameter standardization enables the network to learn consistent geometric transformations across diverse crown shapes, improving prediction accuracy while maintaining ease of operation.

Inventive Principle:

Principle #35Parameter changes

Solution Approach 2:

The patent replaces the traditional mechanical/mathematical crown-morphing system with a data-driven deep learning system. Instead of using mathematical reference models and morphing algorithms, the system uses a trained neural network that has learned the complex mapping between crown and root geometries from training data, achieving higher accuracy without requiring X-ray inputs.

Inventive Principle:

Principle #28Mechanics substitution (Replace mechanical system)

2Measurement precision

If specific 3D reference models are developed for each tooth and patient class, then the accuracy for that specific class improves, but the device complexity and development requirements increase significantly

Engineering Contradiction:

Improveaccuracy for specific tooth/patient classesVSAvoidnumber of reference models required

Core Design Contradiction:

Measurement precisionVSDevice complexity

Solution Approach 1:

The patent creates a universal deep learning model that can handle all tooth types and patient classes simultaneously. The neural network is trained on diverse data including multiple tooth classes (incisors, canines, premolars, molars) and various patient demographics, enabling a single model to generalize across all classes without requiring separate reference models for each category.

Inventive Principle:

Principle #6Universality (Multi-functionality)

Solution Approach 2:

The patent segments the training data into distinct categories (tooth classes, patient age groups, gender) and uses these segmented labels during training to help the network learn class-specific features. The network architecture includes pathways that can adapt to different tooth types while maintaining a unified model structure, avoiding the need for separate models for each segment.

Inventive Principle:

Principle #1Segmentation

3Object-affected harmful factors

If crown 3D image data from optical scanners is used as input, then the system avoids radiation exposure and reduces cost, but the available information for predicting root shape is limited compared to X-ray data

Engineering Contradiction:

Improveradiation exposure to patientVSAvoidinformation available for root shape prediction

Core Design Contradiction:

Object-affected harmful factorsVSLoss of information

Solution Approach 1:

The patent introduces a deep learning neural network as an intermediary that learns the hidden relationship between crown geometry and root shape from training data. This intermediary compensates for the information loss by inferring root characteristics that are not directly observable from the crown, effectively bridging the information gap created by using non-invasive optical scanning instead of X-ray imaging.

Inventive Principle:

Principle #24Intermediary (Mediator)

Data Source

PatentUS11455774B2Automated 3D root shape prediction using deep learning methods

Publication Date: 2022.09.27 PROMATON HLDG BV

AI summary

A computer-implemented method for automated 3D root shape prediction comprising: receiving data defining at least one 3D representation of a tooth and processing the data including: transforming at least part of the data into a voxel representation of a crown; a pre-processor providing the representation of the crown to the input of the neural network trained on clinical representations of real teeth; the first neural network generating a representation of a root or a tooth from the crown, wherein the generation of the representation of the root or tooth includes: determining voxel activations in a voxel space of the output of the deep learning network, each activation representing a probability measure defining the probability that a voxel is part of the root or the tooth; and, determining whether a voxel activation is part of the root or the tooth by comparing the voxel activation with a voxel activation threshold value.