MRI Contrast Dose Simulation via Deep Learning Mapping
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current methods for simulating arbitrary contrast enhancement levels in MRI images rely on physics-based models that are protocol and GBCA-dependent, and deep learning models struggle with limited diverse ground truth data, leading to poor performance in dose reduction and segmentation tasks.
Innovation Solution
A deep learning model-based iterative framework using a global transformer with self-attention and subsampling mechanisms generates images with arbitrary contrast enhancement levels from pre-contrast and post-contrast images, allowing for dose reduction and tumor segmentation by learning a mapping relationship between different dose levels.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If physics-based models are used for simulating arbitrary contrast enhancement levels, then the simulation can be performed without additional training data, but the models are dependent on protocol parameters and GBCA type, reducing their generalizability
Solution Approach 1:
The patent uses a deep learning model to learn the mapping relationship between different contrast enhancement levels by training on simulated images. The model copies the transformation patterns from training data to generate new simulated images at arbitrary dose levels, achieving both generalizability and accuracy without requiring physics-based model dependencies.
Solution Approach 2:
The patent changes the approach from physics-based parameter modeling to data-driven learning. By training the deep learning model on images with known contrast enhancement levels and using iterative optimization, the system can accurately simulate arbitrary contrast levels by adjusting learned parameters rather than relying on fixed physics models.
2Adaptability or versatility
If deep learning models are used for simulating images with arbitrary contrast dose levels, then high capacity and generality are achieved, but the performance heavily depends on the availability of high quality training data
Solution Approach 1:
The patent performs preliminary simulation of images at various contrast enhancement levels using a physics-based framework before training the deep learning model. This preliminary action generates the training data that would otherwise require complex acquisition protocols, allowing the model to learn mapping relationships without needing actual low-dose images for training.
Solution Approach 2:
The patent introduces an intermediary simulation framework that bridges the gap between available data and training requirements. By using a physics-based simulator to generate intermediate images at known contrast levels, the system creates a mediator dataset that enables deep learning model training without requiring direct acquisition of diverse low-dose images.
3Quantity of substance
If standard imaging protocol is modified to acquire various low-dose contrast-enhanced images for training, then high quality training data is obtained, but additional training of MR technicians and operational complexity increase
Solution Approach 1:
The patent uses a physics-based simulation framework to copy and generate virtual images at various contrast enhancement levels. This virtual copying approach replaces the need to actually acquire multiple low-dose images through modified imaging protocols, maintaining training data quality while avoiding the operational complexity of protocol changes.
Solution Approach 2:
The patent substitutes the mechanical process of physical image acquisition with a computational simulation approach. Instead of using modified imaging protocols to capture low-dose images, the system uses a physics-based model to computationally generate these images, eliminating the need for additional MR technician training and protocol complexity.
4Object-affected harmful factors
If minimum dose level is reduced for Gadolinium-based contrast agents, then patient safety and convenience improve, but image quality and diagnostic accuracy may deteriorate
Solution Approach 1:
The patent employs iterative optimization with feedback mechanisms to progressively enhance image quality. The deep learning model iteratively refines the simulated images by comparing them with ground truth images and adjusting the enhancement levels, enabling recovery of fine details and accurate border delineation even at reduced contrast agent doses.
Solution Approach 2:
The patent uses the deep learning model to copy the enhancement characteristics from high-dose images to low-dose images. By learning the transformation patterns from training data, the model can accurately reproduce the contrast enhancement effects at lower doses, maintaining diagnostic accuracy while reducing Gadolinium administration to improve patient safety.
Data Source
AI summary
Methods and systems are provided for simulating images with different dosages. The method comprises: learning a mapping relationship from a post-contrast image to a low-dose image using an iterative method, where learning the mapping comprises generating a plurality of images with intermediate dosages; and applying the mapping relationship to an input images with a higher dose level and a lower dose level to generate one or more simulated images with intermediate dose levels between the higher dose level and the lower dose level.