Deep Learning Medical Imaging Artifact Reduction via Synthetic Training Data
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current medical imaging systems, particularly MRI, face challenges in accurately diagnosing patient conditions with artifacts such as metal implants or fractures, as these distortions can significantly impact the performance of deep learning techniques, and existing methods for simulating patient conditions are inflexible and computationally intensive.
Innovation Solution
A method is introduced to generate training images with simulated medical conditions by blending pathology regions from template source images to target images, using a deep learning network model to improve image classification and artifact reduction, which can be applied across various medical imaging systems.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If deep learning models are trained on real medical images with artifacts, then the models can learn to handle patient conditions, but the training data is limited and expensive to acquire
Solution Approach 1:
The patent creates synthetic training images by copying and modifying real medical images. Specifically, it extracts pathology regions from template source images and blends them onto target images to generate realistic simulated images that replicate various patient conditions and artifacts, thereby expanding training data quantity without requiring additional real scans
Solution Approach 2:
The patent segments medical images into distinct components: it identifies and extracts pathology regions (such as tumors or artifacts) from template images, separates these regions from the background, and then recombines them with target images. This segmentation allows flexible manipulation of specific features while maintaining overall image realism
2Ease of manufacture
If existing methods simulate patient conditions by physically modifying images, then they can create training data, but the methods are inflexible and computationally intensive
Solution Approach 1:
The patent implements dynamic control over the simulation process by allowing flexible adjustment of pathology region parameters (size, shape, position, intensity) and blending characteristics. This dynamic approach enables the system to generate diverse training scenarios adaptively without requiring complex physical modification procedures
Solution Approach 2:
The patent changes key parameters of the simulated pathology regions including their spatial coordinates, dimensions, intensity values, and blending weights. By systematically varying these parameters across multiple simulations, the system generates diverse training data efficiently without requiring complex computational models or physical experiments
3Loss of information
If metal implants are present in MRI scans, then they provide diagnostic information about patient conditions, but they destroy the signal and obscure landmarks
Solution Approach 1:
The patent converts the harmful effect of metal implants (signal destruction and landmark obscuration) into a beneficial training opportunity. By simulating these artifact conditions in controlled training images, the system teaches deep learning models to recognize and compensate for such distortions, thereby improving diagnostic accuracy in real clinical scenarios where artifacts are present
Data Source
AI summary
A medical imaging system includes at least one medical imaging device providing image data of a subject and a processing system programmed to generate a plurality of training images having simulated medical conditions by blending a pathology region from a plurality of template source images to a plurality of target images. The processing system is further programmed to train a deep learning network model using the plurality of training images and input the image data of the subject to the deep learning network model. The processing system is further programmed to generate a medical image of the subject based on the output of the deep learning network model.