Deep Learning MRI Acceleration via Grafted K-Space Artifact Removal
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current MRI examination techniques face challenges in producing high-resolution images for multiple contrasts quickly, as they are affected by motion artifacts and inefficient data processing, leading to structural defects and blurring, which existing deep learning methods struggle to overcome effectively.
Innovation Solution
The method involves acquiring fully sampled reference k-space data and grafting it with partial k-space data to generate a grafted k-space for accelerated examinations, using a deep learning module trained on both datasets to predict and remove artifacts, thereby improving image quality and reducing scan time.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If MRI scan time is reduced by under sampling k-space data, then examination speed is improved, but image quality deteriorates due to structural artifacts and blurring
Solution Approach 1:
A fully sampled reference k-space dataset is acquired before the accelerated scan to capture the subject's anatomical structure. This reference data is prepared in advance to serve as a template for guiding the reconstruction of accelerated images, allowing the system to compensate for the reduced sampling density in subsequent rapid scans.
Solution Approach 2:
A deep learning module acts as an intermediary between the undersampled k-space data and the final image reconstruction. This module processes the accelerated scan data in conjunction with the reference dataset, using learned patterns to generate high-quality images that would otherwise be degraded by the reduced sampling.
2Loss of information
If multiple contrasts are acquired sequentially to ensure comprehensive coverage, then image completeness is improved, but examination time increases
Solution Approach 1:
The deep learning module is trained on multiple contrast types and can process different k-space sampling patterns universally. A single accelerated scan can be reconstructed into multiple contrast images by applying the same deep learning framework with appropriate training data, eliminating the need for separate sequential acquisitions for each contrast type.
Solution Approach 2:
The system changes the reconstruction parameters and processing methods based on the desired contrast type. By adjusting the deep learning module's processing parameters and using contrast-specific training data, the same accelerated scan data can be transformed into different contrast images, reducing the need for multiple physical scans.
3Productivity
If deep learning modules are trained only on accelerated data, then processing speed is improved, but artifact removal capability deteriorates
Solution Approach 1:
The training process merges information from both fully sampled reference datasets and accelerated scan datasets. The deep learning module learns to recognize artifacts by comparing the differences between these two types of data, enabling it to effectively identify and remove artifacts from accelerated images while maintaining processing efficiency.
Solution Approach 2:
The training process incorporates feedback mechanisms where the deep learning module's predictions are continuously refined by comparing reconstructed images with the reference fully sampled images. This feedback loop allows the system to learn from its errors and improve its artifact removal capability over time, ensuring reliable performance even with accelerated data.
Data Source
AI summary
Systems and methods for deep learning based magnetic resonance imaging (MRI) examination acceleration are provided. The method of deep learning (DL) based magnetic resonance imaging (MRI) examination acceleration comprises acquiring at least one fully sampled reference k-space data of a subject and acquiring a plurality of partial k-space of the subject. The method further comprises grafting the plurality of partial k-space with the at least one fully sampled reference k-space data to generate a grafted k-space for accelerated examination. The method further comprises training a deep learning (DL) module using the fully sampled reference k-space data and the grafted k-space to remove the grafting artifacts.