Endoscopic IBD Scoring With Multimodal Machine Learning
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current methods for classifying inflammatory bowel disease (IBD) severity and progression are subjective and suffer from high intrarater and interrater variability, leading to inconsistent scoring and increased costs and inefficiencies in clinical trials.
Innovation Solution
A data processing system utilizing machine learning and computer vision techniques to analyze endoscopic images, combined with electronic health records and omics data, to generate objective and consistent IBD scores by incorporating regression models that account for temporal information and multiple data modalities.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If manual classification scoring is used by physicians or centralized reading platforms, then the process is simple and accessible, but scoring subjectivity and intrarater variability are high leading to inconsistent results
Solution Approach 1:
The patent replaces manual mechanical scoring by physicians with an automated machine learning-based scoring system. The system uses convolutional neural networks to process endoscopic images and generate consistent disease activity scores, eliminating human subjectivity and intrarater variability while maintaining accessibility through automated processing.
Solution Approach 2:
The patent creates a digital copy of the manual scoring process through trained machine learning models. The system learns from annotated training data representing expert physician scoring and replicates this scoring behavior consistently across all patients, ensuring reproducibility and eliminating variability inherent in manual scoring.
2Productivity
If manual scoring methods are used, then the system is simpler, but clinical trials require more patients and take longer due to high variability and inefficiency
Solution Approach 1:
The patent replaces manual scoring operations with automated machine learning processing that can rapidly analyze endoscopic images. This substitution dramatically increases productivity by enabling quick, consistent scoring without requiring additional patient recruitment to compensate for scoring variability, thereby reducing the number of patients needed and accelerating trial timelines.
3Measurement precision
If single still frame image classification is used, then processing is faster, but accuracy degrades in multi-class setups
Solution Approach 1:
The patent applies continuous video frame analysis rather than isolated single-frame classification. The system processes sequential frames from endoscopic videos, maintaining temporal context and accumulating evidence across time to improve scoring accuracy. This continuous processing approach handles the complexity of multi-class IBD classification while managing computation through efficient frame-by-frame analysis.
Solution Approach 2:
The patent performs preliminary frame-level annotations and score assignments during the training phase using ground truth data. These pre-computed annotations are stored and used during inference to guide the classification process, enabling accurate multi-class scoring without requiring exhaustive real-time analysis of every pixel and temporal frame during actual scoring.
Data Source
AI summary
This specification describes systems and methods for performing endoscopy, obtaining medical images for inflammatory bowel disease (IBD) and scoring severity of IBD in patients. The methods and systems are configured for using machine learning to determine measurements of various characteristics related to IBD. The methods and systems may also obtain and incorporate electronic health data of patients along with endoscopic data to use for scoring purposes.