Raw MR Analytics Using AI Without Image Reconstruction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current MR imaging devices require costly hardware due to stringent requirements for homogeneous B0 and B1 fields and linear gradients, and reconstruction methods suppress information based on subjective criteria, which are not optimized for clinical decision-making.
Innovation Solution
An end-to-end artificial intelligence approach directly analyzes raw MR data using machine-learned models to determine clinical findings without image reconstruction, allowing for simpler and less expensive MR hardware configurations with non-homogeneous fields and non-linear gradients.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Loss of information
If image reconstruction is performed using deterministic functions, then images can be generated for radiologist review, but information is suppressed based on subjective criteria not optimized for clinical decision-making
Solution Approach 1:
An AI model is introduced as an intermediary between raw MR data and clinical findings, replacing the traditional image reconstruction step. The AI model directly processes raw data to generate clinical insights, eliminating the information loss associated with deterministic reconstruction algorithms that use subjective criteria.
Solution Approach 2:
The patent extracts and removes the image reconstruction step from the traditional MR imaging workflow. By taking out the reconstruction process that suppresses information, the system directly analyzes raw data to preserve all information content for clinical decision-making.
2Productivity
If Cartesian sampling and FFT reconstruction are used to simplify reconstruction tasks, then reconstruction speed is improved, but information rate of the scan is reduced
Solution Approach 1:
The patent removes the Cartesian sampling and FFT reconstruction pipeline from the system. By eliminating these simplifying assumptions, the system can use alternative sampling strategies that preserve more information while still achieving fast processing through the AI model's direct analysis of raw data.
3Device complexity
If simplifying assumptions are made in reconstruction to reduce hardware requirements, then device complexity is reduced, but hardware performance and cost are adversely affected
Solution Approach 1:
Instead of simplifying hardware to accommodate traditional reconstruction requirements, the patent inverts the approach by using advanced AI processing to compensate for simpler hardware. The system accepts non-homogeneous fields and non-linear gradients that would normally degrade image quality, but the AI model directly processes the raw data from these simpler systems to produce accurate clinical findings, effectively making the hardware requirements less stringent without sacrificing performance.
4Manufacturing precision
If homogeneous B0 and B1 fields and linear gradients are required for traditional MR imaging, then image quality is maintained, but hardware costs increase
Solution Approach 1:
The patent reverses the traditional relationship between hardware precision and image quality. Instead of requiring homogeneous fields and linear gradients to achieve good image quality, the system uses an AI model to directly process raw data from systems with non-homogeneous fields and non-linear gradients, achieving accurate clinical findings without the need for expensive, precision hardware.
Solution Approach 2:
The patent replaces the mechanical/physical requirements for homogeneous fields and linear gradients with an computational approach. The AI model compensates for hardware imperfections through intelligent data processing, substituting the need for precision mechanical field generation with software-based correction and analysis.
Data Source
AI summary
For data analytics in magnetic resonance (MR) scanning, the scanning configuration information and the resulting raw data are directly used to determine the analytics or clinical decision. Artificial intelligence provides a value for a clinical finding characteristic of the patient based on the raw data from scanning and the controls used to scan, allowing the value to be based on all of the information content of the scan results. Reconstruction is not needed, allowing for simpler hardware, such as hardware with less homogeneous B0 and/or B1 fields than the norm and/or non-linear gradients.