Deep Learning Model for Tomosynthesis Image Contrast and Artifact Removal
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Tomosynthesis imaging generates two-dimensional images with limited contrast and missing information due to lower X-ray doses and artifacts from X-ray tube movement, failing to accurately depict mammary glands and calcifications compared to mammography.
Innovation Solution
A medical image processing apparatus that uses a trained model for deep learning to generate high-definition two-dimensional images from tomosynthesis data, incorporating projection data and volume data to match mammography contrast and reduce artifacts, by integrating image acquisition, generation, and learning functions.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If minimum intensity projection (MinIP) is used to generate two-dimensional image from tomosynthesis data, then a single digest image can be provided, but information on mammary glands and contrast are lost
Solution Approach 1:
The patent applies parameter changes by using a trained model that learns optimal transformation parameters from paired tomosynthesis and mammography images. The model adjusts multiple parameters simultaneously to generate two-dimensional images that preserve mammary gland information while maintaining the efficiency of single-image output.
Solution Approach 2:
The trained deep learning model serves as an intermediary between tomosynthesis volume data and final two-dimensional images. This intermediary processes the three-dimensional data through learned transformations to produce images that retain critical diagnostic information like mammary glands and calcifications.
2Object-affected harmful factors
If lower X-ray doses are used in tomosynthesis imaging, then patient radiation exposure is reduced, but image contrast deteriorates
Solution Approach 1:
The patent uses copying by training a deep learning model on pairs of tomosynthesis images (acquired at lower dose) and corresponding mammography images (acquired at higher dose). The model learns to copy the contrast characteristics and anatomical details from the high-quality reference images and apply them to the low-dose tomosynthesis data, generating two-dimensional images with mammography-like contrast.
3Loss of information
If X-ray tube movement is used in tomosynthesis imaging, then three-dimensional volume data can be acquired, but artifacts are introduced
Solution Approach 1:
The patent applies taking out by using the trained model to extract and eliminate motion artifacts from the tomosynthesis data. The model has learned to distinguish between actual anatomical structures and artifacts caused by X-ray tube movement during acquisition, selectively removing artifacts while preserving genuine anatomical information in the generated two-dimensional images.
Data Source
AI summary
A medical image processing apparatus of an embodiment includes processing circuitry. The processing circuitry is configured to acquire medical image data on the basis of tomosynthesis imaging of a test object, and input the acquired medical image data of the test object to a trained model to acquire a two-dimensional image data, the trained model being generated by learning of two-dimensional image data on the basis of X-ray imaging of a person and image data on the basis of tomosynthesis imaging of the person who is subjected to the X-ray imaging.