Out-of-Distribution Testing Neural Network for MRI Reconstruction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current compressed sensing algorithms in MRI face challenges with neural networks when presented with data outside their training distribution, leading to incorrect reconstructions and a lack of efficiency in processing large k-space data, which can result in poor image quality and lengthy reconstruction times.
Innovation Solution
The implementation of an out-of-distribution testing neural network, trained as a discriminator in a generative adversarial network, to identify if undersampled k-space data belongs to the training distribution by analyzing noise and artifacts in the reconstructed images, providing a test signal to determine the feasibility of reconstruction before proceeding with computationally intensive processes.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If neural networks are used for compressed sensing reconstruction, then reconstruction accuracy is improved, but reliability deteriorates when data is outside training distribution
Solution Approach 1:
The system performs a preliminary out-of-distribution test on the undersampled k-space data before executing the full compressed sensing reconstruction. The test neural network evaluates whether the data belongs to the training distribution, and only if the test passes does the system proceed to the computationally intensive reconstruction process. This preliminary action prevents unreliable reconstructions from being generated on out-of-distribution data.
2Measurement precision
If compressed sensing algorithms process large amounts of k-space data, then image quality is improved, but processing time increases
Solution Approach 1:
The system performs a quick out-of-distribution test using a test neural network before initiating the full compressed sensing reconstruction process. This preliminary evaluation determines whether the undersampled k-space data is suitable for reconstruction, allowing the system to skip the time-consuming reconstruction process when the data is likely to fail, thus saving computational time while maintaining image quality standards.
3Measurement precision
If neural networks are trained on specific data distributions, then reconstruction performance is improved, but adaptability deteriorates to out-of-distribution data
Solution Approach 1:
The system introduces an intermediary out-of-distribution test neural network that acts as a mediator between the data acquisition process and the main reconstruction process. This test network evaluates whether the input data belongs to the training distribution before allowing the main reconstruction network to process it. The intermediary test serves as a gatekeeper that maintains the specialized performance of the reconstruction network while adapting to handle out-of-distribution data by preventing processing of unsuitable inputs.
Data Source
AI summary
Disclosed herein is a medical system (100, 300) comprising a memory (110) storing machine executable instructions (120). The medical system further comprises a computational system (104). Execution of the machine executable instructions causes the computational system to: reconstruct or receive (202) a test magnetic resonance image reconstructed from undersampled k-space data; receive (204) a test signal in response to inputting the test magnetic resonance image into an out of distribution testing neural network; and provide (206) the test signal. The test neural network is configured for outputting the test signal in response to receiving the test magnetic resonance image. The test signal is descriptive if the test magnetic resonance image is within a training distribution defined by a set of training data.