Radiological Image Prediction Across Variable Contrast Agent Doses
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing machine learning models are unable to generate radiological images that depict an examination area after the application of different amounts of contrast agent, requiring at least two images for generating images with a standard amount of contrast agent, and are not suitable for predicting images with larger quantities.
Innovation Solution
A machine learning model is trained using a sequence of representations of an examination area after successive applications of increasing amounts of contrast agent, accounting for deviations using an error function to minimize model parameters, enabling predictions based on a single input image.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If a machine learning model is trained to map native image and low-contrast image to full-contrast image, then the model can generate images with standard contrast agent amount, but the model cannot generate images with different or larger contrast agent amounts
Solution Approach 1:
The patent applies dynamics by making the contrast agent amount a variable parameter rather than a fixed target. The model is trained to predict images for multiple different contrast agent amounts (a1, a2, ..., an) rather than a single standard amount, enabling flexible adaptation to various contrast enhancement levels while maintaining a unified model structure
Solution Approach 2:
The patent implements parameter changes by introducing the contrast agent amount as an explicit trainable parameter. The loss function incorporates the contrast agent amount variable, allowing the model to learn the relationship between different contrast agent quantities and their corresponding image representations, thus enabling generation of images with varying contrast levels
2Manufacturing precision
If at least two radiological images (native and low-contrast) are used to generate full-contrast images, then the generation accuracy improves, but the procedure complexity and time increase
Solution Approach 1:
The patent applies preliminary action by pre-training the model on multiple contrast agent amounts during the training phase. This preliminary learning of contrast enhancement patterns across different levels enables the model to directly generate images for any desired contrast amount without requiring multiple input images at inference time, thus reducing processing time while maintaining accuracy
3Adaptability or versatility
If the model is trained on multiple contrast agent amounts, then the model can predict images for larger contrast agent quantities, but the training data requirements and computational resources increase
Solution Approach 1:
The patent implements universality by designing a single unified model that can handle multiple contrast agent amounts and different prediction scenarios (standard amount, larger amounts, intermediate amounts). This multi-functional model eliminates the need for separate models or extensive additional training data for each specific contrast level, as the model learns generalizable contrast enhancement patterns that apply across the full range of contrast agent quantities
Data Source
Figure 1(a)~1(b)
Figure 2(A)~2(D)
Figure 3(A)~3(D)
AI summary
The present invention relates to the technical field of radiology, and in particular to assisting radiologists in radiological examinations using artificial intelligence methods. The present invention relates to training a machine learning model and using the trained model to predict representations of an examination area after applications of different amounts of a contrast agent.