Colon CAD Anomaly Matching Across Multiple Views
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing CAD systems for colon cancer detection face challenges in accurately identifying suspicious polyps across multiple views of the colon, leading to high false positives and missed detections, as they assume anomalies in multiple images are of interest, and many polyps are only detectable in one view.
Innovation Solution
A system and method that register and match candidate suspicious anomalies across multiple images of the colon acquired in different positions or at different times, using feature-based matching and classification to distinguish true positives from false positives, and presenting matched anomalies with distinct notation for improved detection.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If prior art CAD systems assume anomalies appearing in multiple images are of interest, then sensitivity for detecting persistent anomalies is improved, but false positive rate increases because many polyp-like false positives persist across multiple volumes
Solution Approach 1:
The patent inverts the conventional assumption by treating anomalies that appear in only one image as more likely to be true positives, while anomalies persisting across multiple images are treated as suspicious false positives. This inversion of the traditional multi-view assumption resolves the contradiction by flipping the classification logic to reduce false positives while maintaining sensitivity through alternative validation methods.
Solution Approach 2:
The system dynamically adjusts the classification of anomalies based on their persistence across multiple images. Rather than applying a static rule, the system adapts its evaluation criteria, considering the specific characteristics and patterns of each anomaly's appearance across different image sets, allowing flexible differentiation between true positives and false positives.
2Object-generated harmful factors
If prior art CAD systems require anomalies to appear in multiple views for confirmation, then false positives are reduced, but sensitivity decreases because many polyps are detectable in only one view
Solution Approach 1:
The patent inverts the conventional multi-view confirmation requirement by establishing that single-view anomalies can be classified as true positives through alternative validation mechanisms. This inversion allows the system to maintain low false positive rates while capturing polyps that are visible in only one view, thereby resolving the sensitivity loss.
Solution Approach 2:
The system introduces an intermediary classification process that evaluates single-view anomalies using additional criteria and validation methods. This intermediary step acts as a mediator between the strict multi-view requirement and single-view detections, allowing flexible classification that maintains both low false positive rates and high sensitivity.
3Device complexity
If multiple views of the colon are separately analyzed by CAD systems, then processing complexity is reduced, but detection accuracy decreases due to inability to leverage correlations across views
Solution Approach 1:
The patent segments the analysis process into distinct stages: initial separate detection in each view, followed by correlation analysis across views. This segmentation allows the system to maintain low processing complexity in the initial detection phase while improving accuracy through subsequent multi-view correlation, resolving the contradiction between complexity and accuracy.
Solution Approach 2:
The system transitions from analyzing each view independently (2D separate analysis) to incorporating the temporal/dimensional aspect of multiple views by examining anomaly persistence and patterns across the image set. This dimensional expansion improves detection accuracy by leveraging inter-view correlations without proportionally increasing processing complexity.
Data Source
AI summary
Systems, computer-readable media, and methods are presented that identify suspicious anomalies in a colon with higher sensitivity and at a lower false positive rate. A plurality of images of an anatomical colon is acquired. Candidate suspicious anomalies are identified in each image. The candidate suspicious anomalies across images are then compared using registration and matching. Features of candidate suspicious anomalies across images may be jointly evaluated to perform classification.