Parallel MRI Reconstruction With Self-Calibrated Coil Sensitivity
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing parallel MRI methods require ground truth data for supervised training of deep neural networks, making them challenging when ground truth is unavailable, and pre-calibrated coil sensitivity maps can lead to imaging artifacts and inaccurate results.
Innovation Solution
A self-supervised deep learning method for parallel MRI that estimates coil sensitivity maps and reconstructs images without ground truth data, using a coil sensitivity estimation module and an unfolded regularization by denoising module to refine images and enforce data consistency.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If supervised deep learning methods are used for joint image reconstruction and coil sensitivity estimation, then image quality and estimation accuracy are improved, but the requirement for fully sampled ground-truth data increases device complexity and makes application challenging
Solution Approach 1:
The system performs self-calibration by using the acquired undersampled measurements themselves as training data for the deep neural network. The network learns to estimate coil sensitivity maps from the available measurements without requiring external ground truth data, making the system self-sufficient and eliminating the need for separate calibration scans or fully sampled reference data.
Solution Approach 2:
Instead of using ground truth data to train the network and then applying it to undersampled data, the invention inverts the approach by training the network directly on the undersampled measurements to predict both the images and coil sensitivity maps simultaneously. This reversal eliminates the dependency on ground truth data while maintaining estimation accuracy.
2Reliability
If traditional parallel MRI calibration methods are used, then coil sensitivity maps can be obtained, but the calibration process requires additional time and cannot be performed without ground truth data
Solution Approach 1:
The invention merges the image reconstruction process and coil sensitivity calibration into a single unified deep learning framework. Both tasks are performed simultaneously in an end-to-end manner, eliminating the need for separate calibration steps and reducing total processing time while maintaining reliability of the calibration.
Solution Approach 2:
The deep neural network is pre-trained to perform both image reconstruction and coil sensitivity estimation in one integrated model. This preliminary preparation of the network allows it to directly process undersampled measurements and output both reconstructed images and accurate coil sensitivity maps without requiring additional calibration time during actual operation.
3Productivity
If high levels of k-space subsampling are applied to accelerate data acquisition, then productivity is improved, but image quality and reconstruction accuracy deteriorate
Solution Approach 1:
The deep neural network acts as an intermediary that bridges the gap between highly subsampled k-space data and high-quality reconstructed images. The network learns to infer the missing information from the limited measurements by leveraging patterns learned during training, thereby maintaining image quality even at high subsampling rates where traditional methods would fail.
Data Source
AI summary
Systems and methods for image reconstruction for parallel MR imaging are disclosed that receive a k-space single-coil measurement dataset that includes at least two k-space single-coil measurement sets, transforming the k-space single-coil measurement dataset to an estimated CSM using a coil sensitivity estimation module, and transforming the k-space single-coil measurement dataset and the estimated CSM into a final MR image using an MRI reconstruction module. In some aspects, the coil sensitivity estimation module and MRI reconstruction module include deep learning neural networks trained without the use of ground truth data.