Medical Image Processing Using Machine Learning for Scatter Correction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current X-ray imaging techniques, such as cone beam computed tomography (CBCT), face challenges in noise correction and scatter correction due to the need for extensive engineering effort, slow processing times, and inconsistent results, particularly with factors like scatter degrading image quality.
Innovation Solution
A medical image processing method that generates a trained machine learning model using simulated projection and scatter data from 3D training data, allowing for efficient scatter correction and other image enhancements like saturation and truncation correction, noise reduction, and artefact removal through neural networks.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Manufacturing precision
If current noise correction and scatter correction techniques are used, then image quality can be improved, but extensive engineering effort and parameter tuning are required
Solution Approach 1:
The patent creates a digital twin (virtual model) of the imaging system that replicates its physical behavior. This virtual model is trained to predict scatter and noise patterns, replacing complex physical correction processes with a learned digital approximation that requires minimal engineering intervention.
Solution Approach 2:
The patent replaces traditional mechanical/mathematical correction systems with a machine learning-based approach. Instead of using complex algorithms requiring extensive parameter tuning, a trained neural network model automatically performs corrections, substituting the mechanical correction process with an intelligent software system.
2Manufacturing precision
If current correction processes are applied, then some image degradation can be addressed, but processing time becomes very slow
Solution Approach 1:
The patent performs correction operations in advance by training the machine learning model offline on large datasets. Once trained, the model can rapidly apply corrections to new images without requiring complex real-time calculations, thus improving processing speed while maintaining correction quality.
Solution Approach 2:
The virtual model created through digital twin technology pre-learns correction patterns from training data, enabling it to quickly predict and correct scatter and noise in new images without repeating complex iterative calculations, thereby significantly reducing processing time.
3Reliability
If traditional correction methods are used, then correction can be applied, but results are inconsistent
Solution Approach 1:
The digital twin creates a consistent virtual representation of the imaging system that reproduces its behavior across different conditions. This standardized virtual model ensures consistent correction results by eliminating variability introduced by manual parameter tuning and adapting to different imaging scenarios.
Solution Approach 2:
The machine learning model is designed to dynamically adapt to different imaging conditions, patient anatomies, and scan parameters. Rather than relying on fixed correction parameters, the model adjusts its predictions based on the specific characteristics of each image, ensuring consistent quality across diverse scenarios.
Data Source
Figure 1
Figure 2
Figure 3
AI summary
A medical image processing method according to an embodiment is a medical image processing method of applying an X-ray image data (101) acquired by a first radiation apparatus (1300) to a trained machine learning model, and to outputting a high-quality X-ray image data (105) as a processed image. The machine learning model is generated by acquiring, based on 3D data (409) acquired by using a second radiation imaging apparatus (500), simulated projection image data (413) representative of when a object is imaged by the first radiation apparatus (1300) and the simulated data with degraded image quality, and applying a training process to an untrained machine learning model to produce the trained machine learning model by using the simulated projection data (413) and the simulated data with degraded image quality. The first radiation apparatus is an X-ray diagnosis apparatus comprising a C-arm. The second radiation apparatus is a CT (Computed Tomography) apparatus.