Lung and Rib Segmentation in Chest Radiographs Using Hough Transform
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current lung segmentation methods in digital chest radiographs face challenges due to overlapping tissues and uneven density distribution, leading to inaccurate boundaries and missed lung corners, while rib segmentation is hindered by overlapping ribs with lung abnormalities, necessitating improved techniques for precise segmentation.
Innovation Solution
A method involving obtaining a lung model through a training set of chest images, pre-processing images using Gaussian filters and bilateral filtering, and applying generalized Hough transform and dynamic programming algorithms to accurately segment lungs and ribs, with the system capable of aligning and transforming segmentation results back to the original image coordinate system.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Device complexity
If classic lung segmentation methods are used, then the segmentation process is simple, but the segmentation accuracy deteriorates due to zigzag boundaries and missed lung corners
Solution Approach 1:
The patent segments the lung boundary detection process into multiple stages: initial boundary detection using gradient magnitude, followed by dynamic programming optimization to smooth zigzag boundaries, and final refinement using active contours. This multi-stage segmentation approach improves accuracy without excessive complexity.
Solution Approach 2:
The patent performs preliminary actions by first detecting the initial lung boundary using gradient magnitude and Hough transform before applying dynamic programming optimization. This preliminary boundary detection provides a starting point for subsequent refinement, improving final segmentation accuracy.
2Ease of manufacture
If rule-based approaches (threshold segmentation, edge detection) are used, then the method is simple to implement, but segmentation accuracy deteriorates due to low X-ray image quality
Solution Approach 1:
The patent introduces an intermediary active contour model that evolves from the initial boundary detection result. This intermediary model smooths the boundary and resolves ambiguities caused by low image quality, bridging the gap between simple edge detection and accurate segmentation.
Solution Approach 2:
The patent replaces pure rule-based mechanical segmentation with a physics-inspired active contour model that uses energy minimization principles. This substitution allows the boundary to evolve naturally based on image gradients and smoothness constraints, improving accuracy over rigid rule-based methods.
3Measurement precision
If pixel classification approaches (neural networks, genetic algorithms) are used, then segmentation accuracy may improve, but the computational complexity and training requirements increase significantly
Solution Approach 1:
The patent employs an unsupervised active contour approach that automatically adapts to the image characteristics without requiring manual training data or parameter tuning. The model self-adjusts based on the image gradients and boundary smoothness constraints, avoiding the computational burden of training neural networks.
Solution Approach 2:
The patent dynamically adjusts the energy function parameters during the active contour evolution process, balancing the competing forces of gradient attraction and smoothness constraint. This adaptive parameter adjustment achieves accurate segmentation without fixed, pre-trained parameters.
4Device complexity
If ribs are not segmented and removed, then the processing is simpler, but detection accuracy of lung abnormalities deteriorates due to overlapping ribs
Solution Approach 1:
The patent extracts and segments rib structures from the chest radiograph using dedicated rib segmentation algorithms. By separating ribs from the lung fields, the method eliminates rib interference with abnormality detection while maintaining processing efficiency through specialized rib detection routines.
Data Source
AI summary
The present disclosure relates to a method and a system for image segmentation, the technique includes: obtaining a lung image and a lung model based on a plurality of chest image samples in a training set; pre-processing a lung image; acquiring a binary image of boundaries of the lung image; performing the generalized Hough transform on the binary image to locate initial boundaries of the lung image and obtain a Hough location; aligning the lung model to Hough location to obtain an alignment result; applying dynamic programming algorithm to the alignment result to obtain a segmentation result; and transforming the segmentation result back to the original coordinate system.