Adversarial Motion Detection for MRI Scanner Invariance
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current motion detection methods in magnetic resonance (MR) imaging face challenges in accurately detecting motion artifacts across varying MR imagers and contrasts, leading to inconsistent image quality and diagnostic reliability.
Innovation Solution
An adversarially trained regression model, specifically a convolutional neural network, is used to detect motion in MR images, with a discriminator that differentiates between MR imager types and contrasts, enabling the model to learn features invariant to these variations, thus providing a motion severity score and improving quality control.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If a generic convolutional neural network is used for motion detection, then classification results are promising, but the model does not generalize to varying MRI contrasts or MR imagers
Solution Approach 1:
The patent applies domain adaptation by transforming the feature space through adversarial training, changing the parameters of the neural network to be invariant across different domains (contrasts and imagers). The discriminator learns to identify domain-specific features while the generator learns to produce domain-invariant features, resolving the contradiction between accurate motion detection and generalization across varying MRI parameters.
Solution Approach 2:
The patent segments the feature extraction process into domain-specific and domain-invariant components. By using a discriminator to identify domain-specific features (contrast and imager characteristics) and a generator to extract domain-invariant features (motion-related features), the system separates the problematic domain variations from the target motion detection function.
2Reliability
If motion detection is performed with high accuracy across all MR conditions, then diagnostic reliability improves, but the system complexity increases due to need for multiple models
Solution Approach 1:
The patent creates a universal motion detection system that handles multiple MRI contrasts and imager types through a single adversarial training framework. Instead of requiring separate models for each contrast type or imager, the domain-adapted neural network learns to generalize across all conditions, reducing system complexity while maintaining high diagnostic reliability across diverse MR imaging scenarios.
Data Source
AI summary
For detecting motion in MR imaging, a regression model, such as a convolutional neural network, is machine trained. To generalize to MR imagers, MR contrasts, or other differences in MR image generation, the regression model is trained adversarially. The discriminator for adversarial training discriminates between classes of the variation source (e.g., type of MR imager or type of contrast) based on values of features learned in the regression model for detecting motion. By adversarial training, the regression model learns features that are less susceptible or invariant to variation in image source.