Medical Image Conversion Model Using Domain Labels
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing image generation models are limited in generating images in target representation formats other than the format used during training, and may lose unique features during domain conversion, with processing mainly focused on scene recognition rather than versatile medical image transformations.
Innovation Solution
A learning device comprising a first network for deriving subject models, a second network for dimensionally compressing features based on target information, and a third network for generating virtual images in target representation formats, using teacher images and data with diverse formats to train these networks and ensure accurate conversion.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If a conversion model is trained to convert images between specific domains (e.g., CT to MRI), then the conversion accuracy for those specific domains is improved, but the model cannot generate images in target representation formats other than the training format
Solution Approach 1:
The patent applies universality by designing a conversion model that can handle multiple domain conversions (CT to MRI, MRI to CT, and various MRI weighting conversions) using a single trained model. The model accepts domain labels as input and can generate target images in different representation formats beyond what was specifically used during training, making the system versatile across multiple medical imaging domains.
Solution Approach 2:
The patent uses parameter changes by introducing domain labels as controllable parameters that guide the conversion process. By changing the domain label parameter, the model can switch between different conversion tasks (e.g., from CT-MRI conversion to T1-T2 MRI conversion) without retraining, allowing flexible adaptation to different target representation formats.
2Adaptability or versatility
If domain conversion is performed to transform images between different representation formats, then the versatility of image types is improved, but unique features of the input image may be lost
Solution Approach 1:
The patent implements feedback through the discriminator network that evaluates both the authenticity of generated images and the preservation of domain-specific features. The discriminator provides feedback signals during training to guide the generator in maintaining unique features (such as anatomical structures in medical images) while performing domain conversion, thereby preventing information loss.
Solution Approach 2:
The patent applies dynamics by using a dynamic domain label input that allows the model to adaptively adjust the conversion process based on the specific input image and desired target format. The model can dynamically preserve important features while transforming less critical aspects, balancing feature integrity with format versatility.
3Reliability
If existing conversion models are used for specific domain transformations, then the conversion reliability for those domains is improved, but the models are limited to scene recognition processing rather than versatile medical image transformations
Solution Approach 1:
The patent achieves universality by creating a single conversion model that performs multiple medical image transformation tasks including CT to MRI conversion, MRI to CT conversion, and various MRI weighting conversions (T1 to T2, T2 to T1, etc.). This multi-functional model maintains reliability across all these different medical imaging domains while expanding processing scope beyond simple scene recognition.
Data Source
AI summary
An image generation device derives, for a subject including a specific structure, a subject model representing the subject by deriving each feature amount of the target image having the at least one representation format and combining the feature amounts based on the target image. A latent variable derivation unit derives a latent variable obtained by dimensionally compressing a feature of the subject model according to the target information based on the target information and the subject model. A virtual image derivation unit outputs a virtual image having the representation format represented by the target information based on the target information, the subject model, and the latent variable.