Medical image processing apparatus and medical image processing system
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
In medical image processing, sharing a trained model across facilities with different imaging modalities can lead to degraded analysis precision due to differences in image data characteristics, such as noise and clarity, and requires handling of personal information for model generation, which is cumbersome and risky for information leakage.
Innovation Solution
A medical image processing apparatus generates intermediate and simulatively-acquired image data by using generative adversarial networks to align image data characteristics between facilities, allowing a trained model to analyze data without precision loss, without the need to share personal information for model generation.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If a trained model is shared across facilities with different imaging modalities, then model shareability is improved, but analysis precision is degraded due to differences in image data characteristics
Solution Approach 1:
The patent introduces a translation model as an intermediary that converts image data from one modality's characteristic space to another modality's characteristic space. This mediator enables the trained model to process images from different modalities without direct exposure to their unique characteristics, thereby maintaining analysis precision while achieving model shareability across facilities
Solution Approach 2:
The translation model performs parameter transformation by changing the characteristic parameters of image data (such as noise patterns, contrast distribution, and clarity features) from one modality's parameter space to another modality's parameter space. This parameter transformation allows the trained model to receive standardized input regardless of the source modality, resolving the contradiction between shareability and precision
2Manufacturing precision
If personal information is shared for model generation, then model training quality is improved, but information leakage risk increases
Solution Approach 1:
Instead of sharing actual personal patient information for model training, the patent creates translated copies of image data that preserve the essential characteristics needed for training while removing identifiable personal information. The translation model generates synthetic training data that mimics the statistical properties of real patient data without containing sensitive personal identifiers, thus improving training quality while eliminating information leakage risks
Data Source
AI summary
A medical image processing apparatus according to an embodiment is a medical image processing apparatus that generates image data to be used in a trained model trained with image data acquired by a second apparatus, from image data acquired by a first apparatus and that includes a first obtaining unit and a first generating unit. The first obtaining unit is configured to obtain first patient image data obtained by the first apparatus by imaging a patient. The first generating unit is configured to generate, by using a first generative model and from the first patient image data, intermediate image data in which a characteristic of first image data acquired by the first apparatus has been deleted and is configured to generate, by using a second generative model and from the intermediate image data, second patient image data having a characteristic of second image data acquired by the second apparatus.