Scar Segmentation From Low-Dose Contrast MRI Using Neural Networks
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing scar segmentation methods using high doses of gadolinium-based contrast agents in MRI are inaccurate and unreliable due to safety concerns, and reducing the dose leads to inadequate assessment of scar extent.
Innovation Solution
A method for training an artificial neural network using low-dose contrast agent images and reference scar segmentation masks to accurately segment scars, employing techniques like U-Net architecture, image registration, and full width at half maximum method to enhance scar visibility and reliability.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If a full-dose of contrast agent is used for scar segmentation, then the scar assessment accuracy is improved, but the patient safety and material cost deteriorate due to gadolinium accumulation risks
Solution Approach 1:
The patent uses a deep learning neural network to generate a synthesized full-dose contrast agent image from a low-dose image. This copying approach creates a virtual representation of what a full-dose image would look like, allowing accurate scar segmentation without actually administering a full dose of contrast agent to the patient.
Solution Approach 2:
The patent changes the contrast agent dose parameter from full-dose to low-dose for the actual imaging procedure. The deep learning model then compensates for this parameter change by synthesizing the appearance of a full-dose image, enabling accurate segmentation while using reduced contrast agent amounts.
2Object-affected harmful factors
If a low-dose of contrast agent is used for scar segmentation, then the patient safety is improved, but the scar assessment accuracy deteriorates due to insufficient contrast enhancement
Solution Approach 1:
The deep learning neural network creates a synthesized copy of the low-dose image that resembles what a full-dose image would produce. This synthesized image restores the contrast enhancement quality needed for accurate scar segmentation while the actual imaging uses only low-dose contrast agent for patient safety.
Solution Approach 2:
The deep learning model acts as an intermediary between the low-dose input image and the desired full-dose output image. It transforms the low-contrast input into a high-contrast synthesized image, enabling accurate scar assessment without requiring actual full-dose imaging.
3Loss of substance
If traditional segmentation methods are used on low-dose images, then the material cost is reduced, but the segmentation reliability deteriorates due to poor contrast visibility
Solution Approach 1:
Instead of directly segmenting the poor-quality low-dose image, the patent first creates a synthesized full-dose image copy using deep learning. This synthesized image has the contrast visibility needed for reliable segmentation, allowing traditional or automated segmentation methods to work effectively on low-dose acquired data.
Solution Approach 2:
The patent replaces the direct mechanical segmentation process on low-dose images with a two-stage process: first using a deep learning neural network to enhance the image, then performing segmentation. This substitution of the direct segmentation mechanism with an enhanced preprocessing step restores reliability.
Data Source
AI summary
A computer-implemented method for determining scar segmentation includes receiving a medical image of an object to be segmented acquired after an application of a low-dose of contrast agent and determining a scar segmentation mask by applying a trained artificial neural network to the medical image. The low-dose of the contrast agent includes less contrast-agent than a standard full-dose of the contrast agent.