Automated QILD Scoring via Markov Transition Matrix
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current semi-quantitative scoring systems for interstitial lung disease, such as those used in idiopathic pulmonary fibrosis and scleroderma, are unreliable due to expert radiologist dependence and moderate inter-observer variation, limiting their effectiveness in assessing disease status and treatment efficacy.
Innovation Solution
An automated system for generating Quantitative Interstitial Lung Disease (QILD) scores using a combination of de-noising techniques, robust feature selection, classification models, and artificial intelligence, which includes a de-noise algorithm, grid sampling, and a Markov chain transition matrix to calculate transitional scores, reducing variability and improving sensitivity in disease progression assessment.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If visual semi-quantitative scoring by expert radiologists is used, then disease extent can be evaluated, but inter-observer variation reduces reliability
Solution Approach 1:
The patent replaces the manual visual assessment mechanism (radiologist observation and scoring) with an automated image processing system that uses de-noising algorithms, feature extraction, and classification models to objectively quantify interstitial lung disease patterns, thereby eliminating inter-observer variation while maintaining measurement precision
Solution Approach 2:
The system enables self-service by allowing the imaging system itself to perform the scoring function through automated analysis of HRCT images, using embedded algorithms to detect and quantify fibrotic patterns without requiring external expert interpretation, thus improving reliability while preserving measurement capability
2Reliability
If automated image processing with de-noising and classification is used, then scoring reliability is improved, but system complexity increases
Solution Approach 1:
The patent segments the complex image processing task into distinct modular stages: de-noising stage, feature extraction stage, classification stage, and scoring stage. Each stage handles a specific aspect of the analysis, making the overall complex system more manageable and implementable while maintaining high scoring reliability through the coordinated operation of these specialized components
Data Source
Figure 1
Figure 2A~2B
Figure 2C~2D
AI summary
Automated image analysis systems and methods are disclosed to quantify change in fibrosis and interstitial lung disease. The system generates scoring changes in Quantitative Interstitial Lung Disease (QILD) by filtering the uploaded images to minimize cross-site variability within images, sampling from a grid of pixels or voxels within the CT images, classifying individual pixels or voxels within downloaded images based on one or more selected texture features, generating a QILD score for each image based on selected features within the image, and calculating a transition between QILD scores within the plurality of CT images.