Deep Learning Image Quality Assessment for MRI Motion Artifacts
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current image quality assessment systems in medical imaging, particularly in MRI, face challenges in dealing with image artifacts, limited training samples, and variations in clinical personnel's perception of image quality, making it difficult to determine if an image is acceptable for diagnostic purposes.
Innovation Solution
A network architecture that generates motion-corrupted images from original images to predict a motion score, allowing for automated image quality assessment without clinically-annotated data, and uses a generative network to model acceptable image features and a discriminative model to assess image quality, enabling efficient identification of acceptable or unacceptable images.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If conventional image processing techniques and metrics are used to assess image quality, then the system can evaluate image acceptability, but it cannot adequately handle the large variety of image artifacts and encode clinical knowledge
Solution Approach 1:
The patent transforms the image quality assessment from conventional fixed metrics to a deep learning-based parameter transformation system. The neural network learns to map input images to quality scores by automatically adapting to different artifact types and clinical knowledge patterns, resolving the contradiction between reliable assessment and adaptability to various artifacts.
Solution Approach 2:
The patent replaces conventional mechanical image processing techniques with a neural network-based system. This substitution enables the system to handle diverse image artifacts and encode clinical knowledge more effectively, as the neural network can learn complex patterns that conventional metrics cannot capture.
2Measurement precision
If discriminative learning models are used for two-class classification, then the system can determine acceptable or unacceptable images, but it cannot deal with image artifact variations and has limited training samples
Solution Approach 1:
The patent applies data augmentation techniques as a preliminary action to generate synthetic training samples with various artifacts. This allows the model to be trained on diverse artifact variations even when actual annotated data is limited, improving both classification precision and the ability to handle artifact variations.
Solution Approach 2:
The patent uses a neural network architecture that dynamically adapts to different artifact types and variations. The model learns to recognize and classify various artifacts through training, making the classification system dynamic rather than static, thereby handling artifact variations effectively.
3Productivity
If automated image quality assessment is implemented, then operational efficiency is improved, but the system cannot account for differences in image quality perception among different clinical personnel
Solution Approach 1:
The patent creates a universal image quality assessment system that can adapt to different clinical personnel's perception criteria. The neural network is trained to learn quality standards that can be customized or adjusted to match different clinicians' preferences, making the system multi-functional and adaptable rather than rigid.
Solution Approach 2:
The patent incorporates feedback mechanisms where the system can be trained and refined based on clinical personnel's assessments and preferences. This feedback loop allows the automated system to gradually align with human perception variations while maintaining operational efficiency.
4Productivity
If image quality assessment is included in imaging pipeline, then system efficiency is improved by ensuring sufficient quality, but re-acquisition costs and patient movement artifacts remain issues
Solution Approach 1:
The patent performs image quality assessment immediately after image acquisition as a preliminary check before proceeding to full diagnostic workflows. This early detection prevents wasting time on obviously poor quality images and allows for immediate re-acquisition if needed, improving overall system efficiency while minimizing time loss.
Solution Approach 2:
The patent enables rapid quality assessment that can quickly determine whether an image is acceptable or needs re-acquisition. This fast screening process allows the system to skip detailed analysis of poor quality images and quickly identify cases requiring re-scan, reducing overall time loss.
Data Source
AI summary
A system and method includes generation of one or more motion-corrupted images based on each of a plurality of reference images, and training of a regression network to determine a motion score, where training of the regression network includes input of a generated motion-corrupted image to the regression network, reception of a first motion score output by the regression network in response to the input image, and determination of a loss by comparison of the first motion score to a target motion score, the target motion score calculated based on the input motion-corrupted image and a reference image based on which the motion-corrupted image was generated.