Learning-Based Maxillofacial CT Segmentation for Similar Tissues

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Existing image segmentation methods for medical images, particularly in CT scans, struggle with accurately distinguishing tissues with similar CT values, requiring human intervention to account for imaging conditions and individual differences.

Innovation Solution

A segmentation device using a learning model generated from training data to segment biologically important regions in maxillofacial images, such as blood vessels and mandibular canals, without human intervention, and calculate three-dimensional positional relationships, enabling improved accuracy and distance measurements.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Extent of automation

If mathematical segmentation methods based on CT values are used, then segmentation can be performed automatically, but segmentation accuracy deteriorates for tissues with close CT values

Engineering Contradiction:
Improveautomatic segmentationVSAvoidsegmentation accuracy
Core Design Contradiction:
Extent of automationVSMeasurement precision

Solution Approach 1:

The patent introduces a learning model as an intermediary between the input image data and the segmentation output. This learning model, trained on annotated training data, acts as a mediator that captures complex relationships between imaging conditions, tissue characteristics, and segmentation results, thereby improving accuracy while maintaining automation.

Inventive Principle:
Principle #24Intermediary (Mediator)

Solution Approach 2:

The patent performs preliminary training of the learning model using annotated training data before actual segmentation. This preliminary action prepares the model to handle variations in imaging conditions and tissue characteristics, enabling accurate automatic segmentation without requiring manual intervention during the actual segmentation process.

Inventive Principle:
Principle #10Preliminary action

2Measurement precision

If human intervention is used for segmentation determination, then segmentation accuracy improves by considering imaging conditions and individual differences, but productivity deteriorates due to manual work requirements

Engineering Contradiction:
Improvesegmentation accuracyVSAvoidsegmentation efficiency
Core Design Contradiction:
Measurement precisionVSProductivity

Solution Approach 1:

The patent enables the segmentation system to serve itself by automatically learning from training data and performing segmentation without human intervention. The learning model captures the expertise of human operators and applies it automatically, achieving both high accuracy and high productivity by eliminating the need for manual segmentation work.

Inventive Principle:
Principle #25Self-service

3Measurement precision

If learning models are used for segmentation, then segmentation accuracy improves without human intervention, but device complexity increases

Engineering Contradiction:
Improvesegmentation accuracyVSAvoidsystem complexity
Core Design Contradiction:
Measurement precisionVSDevice complexity

Solution Approach 1:

The patent replaces complex manual mechanical processes of human segmentation determination with an automated learning model. The learning model, once trained, provides a systematic and repeatable process that captures the complexity of human expertise without requiring human operators, thereby improving accuracy while the complexity is confined to the training phase rather than the operational phase.

Inventive Principle:
Principle #28Mechanics substitution (Replace mechanical system)

Data Source

PatentEP3806034B1Segmentation device
Publication Date: 2025.10.01 J MORITA MANUFACTURING CORP
  • EP3806034B1 patent drawingFigure 1
  • EP3806034B1 patent drawingFigure 2
  • EP3806034B1 patent drawingFigure 3

AI summary

A learning model provided in a segmentation device is a learning model which is generated using training data such that segmentation data of a biologically important region is output when data of a constituent maxillofacial region is input.