Lung Image Segmentation via Lobe Subdivision for Nidus Accuracy
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current methods for segmenting lung images in CT scans, such as bilateral threshold, watershed, level set, and deep learning methods, face challenges with low segmentation speed and accuracy, particularly for pulmonary symptoms, and often require specific data, leading to incomplete segmentation.
Innovation Solution
A method and system that first perform lung lobe segmentation on CT images to generate updated lung image data, which is then used for nidus segmentation, combining deep learning and threshold algorithms with morphological operations to enhance accuracy and precision.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If traditional segmentation methods (bilateral threshold, watershed, level set, graph cut) are used on CT images, then segmentation can be performed, but segmentation speed is low and accuracy is insufficient
Solution Approach 1:
The patent divides the lung image segmentation task into two sequential stages: first performing lung lobe segmentation to obtain separate lung lobe images, then performing nidus segmentation on each lung lobe image. This multi-level segmentation approach improves overall segmentation accuracy by breaking down the complex task into manageable sub-tasks, while the automated nature of each stage maintains processing efficiency
Solution Approach 2:
The patent performs lung lobe segmentation as a preliminary step before nidus segmentation. By first separating the lung lobes and creating focused regional images, the subsequent nidus segmentation operates on smaller, more manageable regions with reduced complexity, thereby improving accuracy without significantly increasing overall processing time
2Productivity
If deep learning methods are used for nidus segmentation, then segmentation speed improves, but segmentation accuracy is insufficient due to insufficient amount of data
Solution Approach 1:
By dividing the segmentation task into lung lobe segmentation followed by nidus segmentation on regional images, the patent creates more focused training datasets for each stage. This regional approach allows deep learning models to learn more specialized features with sufficient data, improving accuracy while maintaining the speed benefits of automated processing
Solution Approach 2:
The patent applies different segmentation strategies to different regions: lung lobe segmentation at the organ level and nidus segmentation at the lesion level. This localized approach allows each deep learning model to be optimized for its specific task and region, improving overall accuracy while maintaining processing speed through specialized architectures
3Device complexity
If direct nidus segmentation is performed on CT images without lung lobe segmentation, then the process is simpler, but segmentation accuracy is insufficient
Solution Approach 1:
The patent implements a two-stage segmentation process where lung lobe segmentation precedes nidus segmentation. This hierarchical approach improves nidus segmentation accuracy by creating focused regional images that reduce background complexity and highlight relevant features, while the modular design keeps each stage relatively simple and automated
Solution Approach 2:
By performing lung lobe segmentation as a preliminary step, the patent prepares optimized input images for the subsequent nidus segmentation stage. This preliminary processing creates region-specific images with enhanced contrast and reduced interference from surrounding structures, thereby improving accuracy without requiring complex manual intervention
Data Source
AI summary
The present disclosure relates to a method, a system, and a storage medium for segmenting a lung image. The method for segmenting a lung image comprises: obtaining medical image data containing a lung region; performing lung lobe segmentation on the medical image data to generate a plurality of lung lobe data subsets; generating updated lung image data based on one or a plurality of lung lobe data subsets in the plurality of lung lobe data subsets; and performing nidus segmentation on the updated lung image data to generate a segmentation image that identifies a pneumonia nidus.