GAN Medical Image Synthesis for Motion Artifact Correction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional methods for correcting patient motion artifacts in medical imaging, such as digital subtraction angiography, are limited by manual input requirements, computational inefficiencies, and struggle to correct non-translational motion, often introducing new artifacts that obscure the vasculature of interest.
Innovation Solution
A generative adversarial network (GAN) system is employed to enhance medical images by training a generative model to remove motion artifacts, using a discriminative model to classify images without artifacts, and leveraging transfer learning to improve performance on different datasets.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If conventional methods of reducing patient motion effects are used, then motion artifacts can be reduced to some extent, but the methods require manual input or significant computational requirements, preventing real-time motion correction
Solution Approach 1:
The patent replaces manual intervention and conventional computational methods with an automated deep learning system. The generative adversarial network (GAN) automatically learns motion correction transformations from training data, eliminating the need for manual input while achieving real-time processing speeds that conventional methods cannot provide.
Solution Approach 2:
The patent transforms the motion correction approach by changing from manual parameter adjustment to automated neural network parameter learning. The GAN model learns optimal transformation parameters through training on paired images, enabling real-time correction without manual computational requirements.
2Adaptability or versatility
If conventional motion correction methods are applied to non-translational motion, then some correction can be achieved, but the methods struggle to correct non-translational motion present in visceral angiograms
Solution Approach 1:
The patent employs a generative adversarial network that learns complex non-rigid transformations from training data. The model captures non-translational motion patterns (such as respiratory motion, peristalsis) by analyzing the relationship between paired images, enabling accurate correction of motion types that conventional rigid transformation methods cannot handle.
Solution Approach 2:
The GAN system creates synthetic training data and learned transformations that replicate and generalize motion correction patterns. By training on paired images with known ground truth, the model copies and adapts correction strategies for various non-translational motion types, improving versatility and accuracy.
3Productivity
If conventional motion correction methods are used, then processing can be performed, but the methods frequently generate new artifacts that obscure the vasculature of interest
Solution Approach 1:
The patent replaces conventional image processing algorithms with a deep learning-based GAN system. The neural network learns to preserve anatomical structures and vasculature while removing motion artifacts, avoiding the generation of spurious artifacts that plague traditional methods. The model's training on high-quality paired images ensures artifact-free correction.
Solution Approach 2:
The generative adversarial network incorporates a discriminator that provides feedback during training to ensure output quality. The discriminator evaluates generated images against ground truth and real images, guiding the generator to produce artifact-free corrections that accurately represent the underlying anatomy without introducing new artifacts.
4Measurement precision
If traditional subtraction methods are used for DSA, then vasculature can be visualized, but patient motion can render background subtraction difficult, limiting the images' diagnostic quality
Solution Approach 1:
The patent applies motion correction as a preliminary step before performing the subtraction operation. By correcting motion artifacts in advance using the GAN model, the system ensures that subsequent background subtraction can proceed with high accuracy, improving both vasculature visualization and diagnostic quality without the limitations imposed by motion during traditional subtraction processing.
Data Source
AI summary
A computer system is configured to remove motion artifacts in medical images using a generative adversarial network (GAN). The computer system instantiates the GAN having one or more generative network(s) and one or more discriminative network(s) that are pitted against each other to train a generative model and a discriminative model. The training uses a training dataset including a plurality of medical images that are previously classified as without significant motion artifacts for diagnostic purposes. The discriminative model is trained to classify medical images as real or artificial. The generative model is trained to enhance the quality of a medical image and remove motion artifacts by producing a medical image directly from a post-contrast image, without using a pre-contrast mask.