Lung CT COVID-19 Quantification Using AI Tissue Segmentation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing CAD systems for COVID-19 diagnosis lack quantitative metrics, fail to utilize radiomics and machine learning for comprehensive data analysis, and do not accurately classify abnormal findings, limiting physician confidence in diagnosis and characterization of the disease.
Innovation Solution
A computing system employing AI models, including semantic segmentation, multiscale texture signature convolutional neural networks, and multitask regression and classification models, to analyze radiological images, quantify COVID-19 tissue extent, and generate a scattergram for precise characterization.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If existing CAD systems are used for COVID-19 diagnosis, then the system structure is simple, but the measurement precision and reliability of diagnosis are insufficient
Solution Approach 1:
The system segments the diagnosis task into multiple specialized modules: semantic segmentation model for tissue type identification, MTS-CNN model for region identification, radiomics for texture characterization, and MTRC model for pattern classification. Each module processes specific aspects of the radiological data independently before integrating results, thereby improving overall diagnosis precision while managing complexity through modular architecture.
Solution Approach 2:
The system transitions from traditional 2D image analysis to multi-dimensional data integration by incorporating radiomics features (texture, shape, intensity), patient metadata, and temporal sequences. This dimensional expansion enables comprehensive characterization of COVID-19 progression and improves measurement precision through multi-parameter analysis.
2Reliability
If existing CAD systems use only small portions of data for diagnosis, then the processing time is short, but the reliability of diagnosis is reduced
Solution Approach 1:
The system performs preliminary data processing by extracting radiomic features and pre-processing patient data before the actual diagnosis. Feature extraction and data preparation are completed in advance, allowing the core diagnosis algorithm to operate more efficiently on prepared data, thus reducing overall processing time while maintaining comprehensive data utilization for high reliability.
Solution Approach 2:
The system continuously processes and integrates multiple data sources (radiological images, patient metadata, temporal sequences) in a continuous workflow rather than discrete steps. This continuous processing approach ensures all available data is thoroughly analyzed without creating bottlenecks, maintaining both comprehensive data usage and acceptable processing speeds.
3Measurement precision
If existing CAD systems do not use radiomics and machine learning, then the system is easier to operate, but the measurement precision and characterization accuracy are insufficient
Solution Approach 1:
The system implements self-service through automated feature extraction, where the radiomics and machine learning models automatically process radiological images and extract relevant characteristics without requiring manual intervention. The system self-adjusts processing parameters and automatically generates diagnoses, reducing the operational burden on users while maintaining high characterization accuracy through sophisticated algorithms.
4Loss of information
If existing CAD systems do not classify abnormal findings, then the system is simpler, but the loss of information about disease extent and type is increased
Solution Approach 1:
The classification system segments disease characterization into distinct categories: ground glass opacity, crazy paving pattern, consolidation, and other patterns. Each category is handled by specialized processing logic that extracts specific diagnostic information, ensuring comprehensive disease characterization while organizing complexity into manageable classification branches that systematically capture all relevant disease features.
Data Source
AI summary
A method includes receiving one or more radiological images of an anatomy of a patient. The method also includes identifying a boundary of different tissue types in the anatomy of the patient based at least partially upon the one or more radiological images. The method also includes identifying one or more regions within the boundary. The one or more regions include a lung region. The method also includes identifying healthy tissue and COVID-19 tissue in the lung region. The method also includes quantifying an extent of the COVID-19 tissue in the lung region by comparing an amount of the COVID-19 tissue in the lung region to an amount of the healthy tissue in the lung region. The method also includes classifying the extent of the COVID-19 tissue in the lung region into one or more of a plurality of COVID-19 classes.