Capsule Endoscopy Image Classification Using Deep Learning and HMM Error Correction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
The existing methods for analyzing images from capsule endoscopy procedures are time-consuming and labor-intensive, requiring healthcare professionals to review thousands of images to identify pathologies and generate reports, which can take several hours and is tiresome.
Innovation Solution
A system and method utilizing a combination of deep learning neural networks and classical machine learning classifiers to classify and segment images of the gastrointestinal tract, providing classification scores and probabilities, and applying error correction through Modified A Posteriori Probability and Viterbi decoding based on Hidden Markov Models to efficiently identify anatomical landmarks and image locations.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If a reader manually reviews thousands of images to evaluate the procedure and generate a report, then the diagnosis accuracy is improved, but the time consumption and labor intensity increase significantly
Solution Approach 1:
The patent segments the image review task into multiple stages: automated pre-screening by AI algorithms to identify potential pathologies, followed by selective manual review by readers only for suspicious cases. This segmentation reduces the total number of images requiring manual evaluation while maintaining diagnostic accuracy.
Solution Approach 2:
The patent introduces an AI-based image analysis system as an intermediary between the raw images and the human reader. This intermediary performs initial screening, prioritization, and triage of images, presenting only the most relevant cases to the reader for final evaluation, thereby reducing time consumption.
2Measurement precision
If a reader manually reviews all images to ensure thorough evaluation, then the measurement precision is improved, but the productivity decreases
Solution Approach 1:
The patent applies partial action by having the AI system perform comprehensive automated analysis of all images, while human readers perform selective verification only on flagged cases. This division ensures thorough evaluation through AI's complete coverage while maintaining high productivity through reduced manual workload.
Solution Approach 2:
The patent replaces the mechanical process of manual image-by-image review with an automated AI-based image analysis system. This substitution maintains evaluation precision through advanced algorithms while dramatically increasing productivity by processing images at machine speed.
3Loss of information
If the system processes and analyzes all captured images, then the completeness of analysis is improved, but the computational resources and time required increase
Solution Approach 1:
The patent applies preliminary action by having the AI system perform rapid pre-analysis of all images to identify and flag potential pathologies before presenting them for detailed review. This preliminary screening ensures no information is lost while reducing the time required for subsequent detailed analysis by focusing only on relevant cases.
Solution Approach 2:
The patent applies local quality by allocating different levels of analysis depth to different images based on their risk profile. High-priority images with suspected pathologies receive comprehensive detailed analysis, while normal images receive streamlined processing, optimizing the balance between completeness and processing time.
Data Source
AI summary
A method for image classification includes accessing a plurality of images of at least a portion of a gastrointestinal tract (GIT) captured by a capsule endoscopy device and for each image of the plurality of images: providing a classification score for each segment of a plurality of consecutive segments of the GIT by a deep learning neural network, and providing a classification probability for each segment of the plurality of consecutive segments of the GIT based on the classification scores by a classical machine learning classifier. The method further includes determining a classification for each image to one segment of the plurality of consecutive segments of the GIT based on processing a signal corresponding to the classification probabilities of the plurality of images.