MR Image Enhancement Training With Cropped Point Spread Functions
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing machine learning models for magnetic resonance (MR) image reconstruction require substantial memory due to the size of training images, which cannot be cropped in undersampled directions, leading to high memory and runtime requirements.
Innovation Solution
Cropping the point spread function in position space and using its Fourier transform for training, allowing for smaller-sized training MR images and reducing memory requirements.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If training images are used at full size to maintain image quality and capture all spatial information, then reconstruction accuracy is improved, but memory requirements and runtime increase substantially
Solution Approach 1:
The patent segments the training images by cropping them to smaller spatial regions while maintaining the essential local image characteristics. This allows the model to learn from representative patches without requiring the entire large image, thereby reducing memory requirements while preserving reconstruction accuracy for the targeted regions.
Solution Approach 2:
The patent applies local quality by focusing training on specific spatial regions with cropped images rather than uniformly processing the entire image. This approach maintains high image quality and reconstruction accuracy in the regions of interest while reducing the overall data volume required for training, thus lowering memory requirements.
2Quantity of substance
If training images are cropped to reduce size and memory usage, then memory requirements are reduced, but the ability to capture long distance correlations and maintain image quality deteriorates
Solution Approach 1:
The patent applies partial action by cropping training images to smaller regions that are sufficient for capturing local correlations and image characteristics. This partial approach provides enough information for the model to learn effective reconstruction patterns without requiring the complete image, thus reducing memory usage while maintaining adequate image quality for training purposes.
3Measurement precision
If full-sized training images are used to capture all spatial information including long distance correlations, then reconstruction accuracy is improved, but training runtime increases
Solution Approach 1:
The patent segments training images into smaller cropped regions, which reduces the computational load during training. By processing smaller image patches instead of full-sized images, the model achieves comparable reconstruction accuracy for local regions with significantly reduced training runtime and computational resource requirements.
Solution Approach 2:
The patent focuses training on local spatial regions through cropping, which reduces the amount of data processed in each training iteration. This local quality approach maintains reconstruction accuracy for the regions of interest while substantially reducing training runtime compared to processing entire full-sized images.
Data Source
AI summary
Techniques are provided for training a machine learning model (MLM) for image enhancement for use in a magnetic resonance (MR) image, in which a point spread function for undersampled MR data acquisition is received. A cropped point spread function is determined, which is given by the point spread function within a predefined spatial region. At least one training MR dataset corresponding to at least one coil channel is received, and a ground truth reconstructed MR image corresponding to the at least one training MR dataset is received. The MLM is trained in a supervised manner depending on the at least one training MR dataset, on the ground truth reconstructed MR image, and on a Fourier transform of the cropped point spread function.