X-Ray to Regional DRR Conversion for Anatomical Structure Separation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing medical imaging conversion methods using generative adversarial networks (GANs) fail to effectively isolate and separate anatomical structures in digitally reconstructed radiographs (DRRs) due to overlapping organs, leading to loss of useful signal and complex pixel segmentation processes.
Innovation Solution
A method using a single generative adversarial network (GAN) or convolutional neural network (CNN) to transform real x-ray images into regional DRRs, isolating each anatomical structure by training on 3D volumes, optimizing both translation and structure separation functions.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Ease of operation
If a single GAN transforms a real x-ray image into a global DRR including all organs, then the conversion process is simple, but anatomical structures cannot be isolated and overlapping organs cannot be separated
Solution Approach 1:
The patent divides the single global DRR generation into multiple regional DRR generations, where each DRR corresponds to a specific anatomical structure or region. The GAN is trained to accept region-of-interest masks as input and generate corresponding regional DRRs, enabling separation of overlapping organs while maintaining conversion simplicity through automated processing
Solution Approach 2:
The patent applies different processing qualities to different regions by generating separate DRRs for each anatomical structure of interest. Each regional DRR is optimized to represent its specific structure with high precision, allowing detailed analysis of individual organs without interference from overlapping structures
2Manufacturing precision
If pixel segmentation is performed on a global DRR using another dedicated CNN, then structure separation is attempted, but the process becomes complex and useful signal is lost due to overlapping organs
Solution Approach 1:
The patent merges the translation function (x-ray to DRR conversion) and structure separation function into a single GAN-based processing step. The GAN is trained to simultaneously perform both functions by incorporating region-of-interest masks as input, eliminating the need for separate segmentation steps and reducing processing complexity
Solution Approach 2:
The GAN model is designed with multi-functionality, serving both as a translation network (converting x-ray images to DRRs) and as a structure separation network (isolating specific anatomical structures). This universal approach consolidates multiple functions into a single system, reducing overall process complexity
3Manufacturing precision
If a single GAN is used to convert x-ray images to regional DRRs representing specific anatomical structures, then structure isolation is achieved without loss of useful signal, but the GAN requires complex training on 3D volumes
Solution Approach 1:
The patent performs preliminary training of the GAN on 3D volume data with pre-segmented anatomical structures and corresponding region-of-interest masks. This preliminary action prepares the model to accurately generate regional DRRs during actual use, ensuring high precision while managing training complexity through systematic data preparation and model development
Data Source
AI summary
Disclosed is a medical imaging conversion method, automatically converting: at least one or more real x-ray images of a patient, including at least a first anatomical structure of the patient and a second anatomical structure of the patient, into at least one digitally reconstructed radiograph (DRR) of the patient representing the first anatomical structure without representing the second anatomical structure, by a single operation using either one convolutional neural network (CNN) or a group of convolutional neural networks (CNN) which is preliminarily trained to, both or simultaneously: differentiate the first anatomical structure from the second anatomical structure, and convert a real x-ray image into at least one digitally reconstructed radiograph (DRR).