Capsule Endoscopy Lesion Detection Using Transfer-Learned CNNs
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current methods for detecting and classifying small bowel and colonic ulcers and erosions in Crohn's capsule endoscopy are time-consuming, prone to human error, and lack accuracy, hindering effective clinical diagnosis and treatment.
Innovation Solution
A deep learning-based method using transfer learning and semi-supervised learning to identify and classify ulcers and erosions in Crohn's capsule endoscopy images, employing convolutional neural networks with frozen feature extraction layers and tailored classification components, optimized through k-fold patient-stratified training and validation.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If manual examination of Crohn's capsule endoscopy images is performed by physicians, then accurate detection and characterization of ulcers and erosions can be achieved, but the process becomes significantly time-consuming and prone to human error
Solution Approach 1:
The patent replaces the manual mechanical examination process with an automated deep learning-based image analysis system. The system uses trained neural networks to automatically detect, classify, and characterize ulcers and erosions in capsule endoscopy images, eliminating the need for time-consuming manual review while maintaining or improving detection accuracy through consistent algorithmic application.
Solution Approach 2:
The system enables self-service by allowing the image analysis to be performed automatically without requiring physician intervention for each image review. The deep learning model independently processes images, identifies lesions, and generates diagnostic reports, freeing physicians from routine examination tasks while preserving accurate detection capabilities.
2Productivity
If deep learning networks are used for automatic lesion identification, then examination time is reduced and productivity increases, but the system requires extensive training data and complex model architecture
Solution Approach 1:
The patent applies transfer learning by pre-training deep learning models on large datasets of endoscopic images before fine-tuning them for specific ulcer and erosion detection tasks. This preliminary training on general medical imagery provides the model with foundational features, reducing the amount of task-specific training data needed and simplifying the overall model architecture while maintaining high productivity in lesion identification.
Solution Approach 2:
The system segments the deep learning approach into distinct modular components: feature extraction networks, classification modules, and post-processing analysis. This segmentation allows each component to be optimized independently, reducing overall model complexity while enabling fast parallel processing that maintains high productivity in lesion identification.
3Device complexity
If traditional image recognition methods are used, then the system is simpler to implement, but it lacks the accuracy and reliability needed for precise lesion classification
Solution Approach 1:
The patent implements a multi-stage classification approach where the system performs initial lesion detection, then applies specialized classification networks for different lesion types (ulcers, erosions, other findings). This partial application of increasingly complex analysis only where needed maintains overall system simplicity while achieving high classification reliability through targeted deep learning deployment.
Solution Approach 2:
The system introduces intermediate processing layers that bridge simple image input and complex classification output. These intermediaries include feature extraction stages, lesion candidate identification, and confidence scoring mechanisms that simplify the overall system architecture while improving classification reliability by breaking down the complex task into manageable sequential steps.
Data Source
AI summary
The present invention relates to a computer-implemented method capable of automatically detecting small bowel and colonic ulcers and erosions in Crohn's capsule endoscopy image data, by classifying pixels as lesion or non-lesion, using a convolutional image feature extraction step followed by a classification step and indexing such lesions in the set of one or more classes.