Polyp Video Detection Using Temporal ROI Aggregation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for detecting polyps from video sequences have limitations in detection performance, particularly in terms of sensitivity and specificity, necessitating improvements to increase the detection rate.
Innovation Solution
A method involving region-of-interest extraction, description using texture and luminance descriptors, classification with fuzzy trees and forests, motion estimation, and image aggregation to enhance polyp detection accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If conventional polyp detection methods are used, then the detection process is simple, but the detection rate (sensitivity and specificity) is insufficient
Solution Approach 1:
The detection process is segmented into distinct stages: region of interest extraction using shape detection, description using multiple descriptors (texture, luminance, color), classification using fuzzy trees/forests, and temporal follow-up. This segmentation allows each stage to be optimized independently while working together to improve overall detection rate
Solution Approach 2:
The method transitions from simple single-frame detection to multi-dimensional analysis by incorporating temporal information through follow-up of regions of interest across successive images. This adds a time dimension to the detection process, enabling better differentiation between true polyps and false positives through temporal consistency analysis
2Measurement precision
If multiple descriptors and classification methods are applied, then the detection rate improves, but the processing time increases
Solution Approach 1:
Regions of interest are extracted and preliminary classification is performed on the first image before temporal follow-up begins. This preliminary action filters out obvious false positives early, reducing the amount of data that needs to be processed in subsequent temporal analysis stages
Solution Approach 2:
The method maintains continuous processing by repeatedly applying description and classification steps (steps b and c) to successive images for regions of interest that pass the initial threshold. This continuous action ensures that temporal patterns are captured while managing processing load through iterative refinement
Data Source
AI summary
The disclosure relates to a method for detecting polyps from a video sequence comprising a plurality of images. The method includes, after an extraction of regions of interests likely to contain a polyp within the different images, a description of said regions and a classification of said regions as likely to contain a polyp or not. The method includes a first aggregation of same regions of interest on said images consisting of maintaining as a region of interest belonging to the first class on a given image, a region of interest classified in the first class for each successive image; and then a second aggregation of images consisting of maintaining as a region of interest on any image comprised between first and second images, the region of interest appearing for the first time on said first image and for the last time on said second image.