3D Flexible Tube Detection in Colonoscopy via Probabilistic Boosting Trees
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current polyp detection methods in colonoscopy suffer from high false positive rates due to the presence of flexible rectal tubes, which are not effectively handled by existing detection methods that rely on rigid template matching and 2D slice tracking, and lack control against overfitting.
Innovation Solution
A hierarchical modeling system using Probabilistic Boosting Trees to detect and segment 3D flexible tubular structures, employing a voting strategy to identify candidate tube parts and a dynamic programming algorithm to combine them into a flexible tube model, reducing false positives by classifying short and long tubes based on shape and appearance models.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Ease of manufacture
If template matching and 2D slice tracking are used for rectal tube detection, then the detection method is simple to implement, but the false positive rate is high due to inability to handle flexible tube shapes
Solution Approach 1:
The patent segments the rectal tube detection problem into multiple components: detecting tube cross-sections in 2D slices, tracking these sections through the volume, and reconstructing the 3D tube structure. This segmentation allows the system to handle flexible shapes by processing manageable 2D components that are then assembled into the complete 3D structure, reducing false positives while maintaining implementation feasibility
Solution Approach 2:
The patent transitions from 2D slice-based detection to 3D volume-based detection by tracking tube cross-sections through multiple slices. This dimensional escalation enables the system to capture the flexible 3D morphology of rectal tubes that cannot be represented by rigid 2D templates alone, thereby reducing false positives while building upon the simple 2D detection foundation
2Measurement precision
If Massive Trained Artificial Neural Network is used to distinguish polyps from rectal tubes, then the distinction accuracy may improve, but the control against overfitting is reduced
Solution Approach 1:
The patent employs dynamic programming to optimally connect detected tube cross-sections through the 3D volume. This dynamic approach adapts to the flexible variations in rectal tube morphology without requiring excessive training data, achieving accurate distinction between tubes and polyps while maintaining control against overfitting through the structured optimization framework
Solution Approach 2:
The patent replaces the Massive Trained Artificial Neural Network approach with a combination of geometric modeling and dynamic programming. This substitution achieves comparable or superior distinction accuracy while providing explicit control against overfitting through the mathematical optimization framework, avoiding the black-box nature of large neural networks
3Device complexity
If 2D slice tracking is used for tube detection, then the computational complexity is reduced, but the accuracy is insufficient because the tube is not always perpendicular to the axes
Solution Approach 1:
The patent segments the 3D tube detection problem into 2D cross-section detection tasks in each slice, followed by tracking these segments through the volume. This segmentation allows the use of simple 2D detection algorithms while achieving accurate 3D tube localization even when tubes are not perpendicular to slice axes, by reconstructing the complete 3D structure from multiple 2D components
Solution Approach 2:
The patent enhances 2D slice-based detection by adding the third dimension through tracking cross-sections through multiple slices. This dimensional escalation enables accurate detection of arbitrarily oriented tubes while building upon computationally efficient 2D detection methods, achieving both low complexity and high accuracy
Data Source
AI summary
The present invention is directed to a system and method for populating a database with a set of image sequences of an object. The database is used to detect a tubular structure in the object. A set of images of objects are received in which each image is annotated to show a tubular structure. For each given image, a Probabilistic Boosting Tree (PBT) is used to detect three dimensional (3D) circles. Short tubes are constructed from pairs of approximately aligned 3D circles. A discriminative joint shape and appearance model is used to classify each short tube. A long flexible tube is formed by connecting all of the short tubes. A tubular structure model that comprises a start point, end point and the long flexible tube is identified. The tubular structure model is stored in the database.