AI Image Generation Model for Non-Gated CT Clinical Parameter Accuracy

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Current methods for calculating clinical parameters from medical images, such as coronary artery calcification scores, using non-electrocardiogram gated CT images are inaccurate due to image blurriness caused by heart movement, and require complex and costly imaging systems, leading to increased radiation dose and subject load.

Innovation Solution

An information processing apparatus and method that uses a trained image generation model to convert non-electrocardiogram gated CT images into pseudo electrocardiogram gated CT images, allowing for accurate calculation of clinical parameters like CAC scores using a simpler and less costly imaging protocol with reduced radiation and subject load.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Object-affected harmful factors

If non-electrocardiogram gated CT imaging is used, then radiation dose and subject load are reduced, but image quality deteriorates due to heart movement blur

Engineering Contradiction:
Improveradiation doseVSAvoidimage quality
Core Design Contradiction:
Object-affected harmful factorsVSMeasurement precision

Solution Approach 1:

The patent creates a pseudo electrocardiogram gated CT image by copying and transforming the non-electrocardiogram gated CT image through deep learning. The image generation model learns the mapping relationship between non-gated and gated images, generating a synthetic gated image that preserves the diagnostic quality while avoiding the need for actual electrocardiogram gated imaging, thus reducing radiation dose while maintaining image quality.

Inventive Principle:
Principle #26Copying

2Measurement precision

If electrocardiogram gated imaging is used, then image quality is improved, but device complexity and cost increase

Engineering Contradiction:
Improveimage qualityVSAvoidimaging system complexity
Core Design Contradiction:
Measurement precisionVSDevice complexity

Solution Approach 1:

The patent replaces the mechanical and procedural complexity of electrocardiogram gated imaging with an artificial intelligence-based image generation system. Instead of requiring synchronized mechanical triggering with the electrocardiogram cycle and complex timing coordination, the system uses a deep learning model to synthetically generate the gated image from a simple non-gated scan, substituting computational processing for mechanical coordination.

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

3Ease of operation

If non-electrocardiogram gated CT imaging is used, then imaging process is simplified, but clinical parameter calculation accuracy deteriorates

Engineering Contradiction:
Improveimaging process simplicityVSAvoidclinical parameter accuracy
Core Design Contradiction:
Ease of operationVSMeasurement precision

Solution Approach 1:

The patent introduces an intermediary step of generating a pseudo electrocardiogram gated CT image from the non-gated image. This intermediate generated image serves as a bridge, allowing the system to maintain the simplicity of non-gated imaging acquisition while achieving the accuracy of gated imaging for clinical parameter calculation, such as coronary artery calcification scoring.

Inventive Principle:
Principle #24Intermediary (Mediator)

Data Source

PatentEP4312229A1Information processing apparatus, information processing method, program, trained model, and learning model generation method
Publication Date: 2024.01.31 FUJIFILM CORP
  • EP4312229A1 patent drawingFigure 1~2
  • EP4312229A1 patent drawingFigure 3
  • EP4312229A1 patent drawingFigure 4

AI summary

An information processing apparatus includes one or more processors, and one or more storage devices that store a program including an image generation model trained to generate, from a first image, a second image that imitates an image obtained by an imaging protocol different from an imaging protocol of the first image. The image generation model is a model trained, through machine learning using training data in which a training image captured by a first imaging protocol is associated with a correct answer clinical parameter calculated from a corresponding image captured by a second imaging protocol different from the first imaging protocol for the same subject as the training image using a modality of the same type as a modality used to capture the training image, such that a clinical parameter calculated from a generation image output by the image generation model approaches the correct answer clinical parameter.