MRI K-space Abnormality Detection via Machine Learning
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current MRI technologies face challenges in efficiently detecting abnormalities from k-space data due to long scan times and high costs, as well as reduced detection accuracy when using sparse data with noise and artifacts, and machine learning methods require high-quality images for input.
Innovation Solution
A system and method that utilize machine learning classification models to detect abnormalities directly from MRI k-space data without reconstructing images, using optimized sampling trajectories and acquisition parameters, and training models on sparse data to achieve accurate detection.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If MRI scanners acquire complete sets of high-quality images for radiologists to read, then detection accuracy is improved, but scan time and cost increase significantly
Solution Approach 1:
The patent extracts only the essential information needed for abnormality detection from the full k-space data, rather than reconstructing complete images. By applying detection models directly to a small portion of raw MR data in k-space domain, the system obtains detection results without forming full images, thereby reducing scan time while maintaining detection accuracy.
Solution Approach 2:
The patent introduces detection models as an intermediary between raw MR data and diagnostic conclusions. These models process k-space data directly to identify abnormalities, eliminating the need for complete image reconstruction and radiologist review, thus achieving fast and accurate detection simultaneously.
2Extent of automation
If machine learning methods are used to detect abnormalities from MRI images, then radiologist intervention is reduced, but complete high-quality images must still be formed which requires long scan time
Solution Approach 1:
The patent extracts only the necessary k-space data portion required for automated detection, rather than acquiring complete image sets. The detection models operate directly on this extracted data in k-space domain, achieving automated abnormality detection without the time penalty of forming complete images.
Solution Approach 2:
The patent shifts the detection process from the image domain to the k-space domain, operating in a different dimensional space. By applying detection models directly to k-space data rather than reconstructed images, the system achieves automated detection with reduced data acquisition time.
3Productivity
If parallel imaging methods and compressed sensing are used to reduce scan time by partially acquiring k-space, then scan time is reduced, but noise and artifacts increase significantly
Solution Approach 1:
The patent changes the approach from attempting to reconstruct high-quality images from sparse data to directly detecting abnormalities from the sparse k-space data itself. By adjusting the detection strategy to work with sparse data characteristics rather than trying to overcome them through reconstruction, the system maintains reliability while achieving high scan speed.
Solution Approach 2:
The patent converts the limitation of having sparse, noisy k-space data into an advantage by designing detection models that operate directly on this data. Instead of viewing sparse sampling as a problem that requires complex reconstruction, the system uses the sparse data directly for detection, turning the apparent harm into a beneficial simplification.
Data Source
AI summary
A system and method to detect abnormality of subjects directly from MRI k-space data are provided. The system includes: at least one computer hardware processor, at least one non-transitory computer-readable storage medium, and at least one computer program stored in the at least one non-transitory computer-readable storage medium and executable on the at least one computer hardware processor, wherein the at least one computer program includes: an acquisition module, configured to obtain target MRI k-space data by scanning a subject, wherein the target MRI k-space data are fully-sampled or undersampled or sparse MRI k-space data; a detection module, configured to obtain and output detection outcome from the target MRI k-space data using detection models; and a model training module, configured to train the detection models based on training data. Hence, the MRI scan time and related cost are reduced, and the accuracy of the detecting results is increased.