Radiation Image Irradiation Field Extraction via Neural Network and Hough Transform
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing image processing technologies struggle to accurately extract the irradiation field in radiation imaging, particularly when the field of interest includes structures with strong edge components like bones or implants, and neural network-based methods face difficulties in classification due to comprehensive feature reliance.
Innovation Solution
An image processing apparatus and method that employs a neural network for preprocessing and contour extraction, using a combination of grid removal, scattered ray reduction, noise normalization, and Hough transform to accurately identify and extract the irradiation field by distinguishing between the field of interest and the collimator region, even in the presence of strong edge components.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If edge intensity-based contour extraction is used to extract the irradiation field, then the extraction process is simple and fast, but structures with strong edge components (bones, implants) cannot be distinguished from collimator contours
Solution Approach 1:
The patent segments the image processing into multiple stages: first using a neural network to generate probability maps for different regions (collimator, irradiation field, body), then using edge detection only within the predicted collimator region. This segmentation allows simple edge-based methods to work effectively by limiting their search space to regions already identified as collimator areas by the neural network.
Solution Approach 2:
The patent introduces probability maps as an intermediary between the neural network output and the final contour extraction. The probability maps provide graded confidence levels for each pixel belonging to different regions, which then guides the edge detection process to focus on high-probability collimator regions while avoiding false edges in the irradiation field.
2Reliability
If neural network-based irradiation field extraction is used, then comprehensive features of multiple images can be determined, but it becomes difficult to classify regions accurately when relying on the neural network alone
Solution Approach 1:
The patent merges two different approaches: neural network-based probability estimation and traditional edge detection algorithms. The neural network provides region classification while edge detection refines the boundaries. By combining these methods, the system achieves more accurate irradiation field extraction than either method could achieve alone, reducing misclassification of structures with strong edges.
3Loss of information
If the irradiation field includes structures with strong edge components, then diagnostic information is preserved, but the structures are misclassified as collimator contours
Solution Approach 1:
The patent applies different processing qualities to different regions: the neural network generates probability maps with high sensitivity for region identification, while edge detection is applied with high precision only within predicted collimator regions. The irradiation field regions, even with strong edges, are protected from aggressive edge detection by the probability map guidance, preserving diagnostic information while avoiding misclassification.
Data Source
Figure 1A~1B
Figure 2A~2B
Figure 3A~3B
AI summary
An image processing apparatus configured to extract an irradiation field from an image obtained through radiation imaging, comprises: an inference unit configured to obtain an irradiation field candidate in the image based on inference processing; a contour extracting unit configured to extract a contour of the irradiation field based on contour extraction processing performed on the irradiation field candidate; and a field extracting unit configured to extract the irradiation field based on the contour.