Medical Image GAN Synthesis for Non-Translational Motion Correction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional methods for reducing patient motion artifacts in medical images, such as digital subtraction angiography (DSA), are limited by the need for manual input, significant computational requirements, and inability to correct non-translational motion effectively, leading to decreased diagnostic quality and increased healthcare costs.
Innovation Solution
A computer system utilizing a generative adversarial network (GAN) is employed to remove motion artifacts from medical images. The system includes a generative network and a discriminative network that are trained on datasets of medical images with and without motion artifacts, allowing for the enhancement of images with significant motion artifacts to achieve diagnostic quality similar to images without motion artifacts.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If conventional methods are used to reduce motion artifacts, then manual input or significant computational requirements are needed, but real-time motion correction is prevented
Solution Approach 1:
The patent replaces conventional mechanical/image-processing-based motion correction methods with a deep generative adversarial network (GAN) system. The GAN consists of a generator network that synthesizes motion-corrected images from motion-artifact-containing images and a discriminator network that evaluates image quality. This substitution enables real-time motion correction by using learned patterns from training data rather than computationally intensive conventional algorithms.
2Adaptability or versatility
If conventional motion correction methods are applied, then translational motion can be addressed, but non-translational motion in visceral angiograms cannot be corrected
Solution Approach 1:
The patent changes the approach from parameter-based motion correction (assuming specific motion models) to data-driven correction using GANs. The generator network learns to map motion-artifact-containing images to motion-corrected images by analyzing patterns from training data, enabling it to handle various motion types including non-translational motions in visceral angiograms without requiring explicit motion models.
3Object-affected harmful factors
If conventional motion correction is performed, then some motion artifacts are reduced, but new artifacts are generated that obscure the vasculature
Solution Approach 1:
The patent implements a feedback mechanism through the discriminator network, which continuously evaluates the generated images and provides feedback to the generator network. This adversarial feedback loop ensures that the generator produces images that not only reduce motion artifacts but also maintain anatomical accuracy and do not introduce new artifacts, as the discriminator penalizes generations that fail to meet quality criteria.
4Measurement precision
If repeated procedures are performed to obtain diagnostic quality images, then diagnostic accuracy is improved, but healthcare costs and procedure times increase
Solution Approach 1:
The patent performs preliminary action by using the GAN system to correct motion artifacts in real-time during the imaging procedure, rather than requiring repeated procedures. The generator network processes images as they are acquired, producing motion-corrected output that maintains diagnostic quality, thereby eliminating the need for repeat procedures and reducing overall procedure time.
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.