Source-linked AI summary
Severity Assessment of Coronavirus Disease 2019 (COVID-19) Using Quantitative Features from Chest CT Images
Zhenyu Tang, Wei Zhao, Xingzhi Xie, Zheng Zhong, Feng Shi, Jun Liu, Dinggang Shen
TL;DR
Manual COVID-19 severity assessment is labor-intensive and may delay patient isolation and treatment, while automatic assessment using chest CT remains limited. This paper builds a random forest model for severity assessment and reports promising performance, while identifying quantitative features related to severity.
Problem
Manual COVID-19 severity assessment is labor-intensive and may delay patient isolation and treatment, while few automatic chest-CT-based assessment methods exist.
Method
A random forest model is built for COVID-19 severity assessment using quantitative features from chest CT images.
Results
0.875 accuracy and 0.91 AUC were achieved, with TPR of 0.933 and TNR of 0.745 on testing samples.
Takeaways & Limitations
The resulting model provides automatic non-severe-versus-severe assessment, and quantitative features related to COVID-19 severity were revealed.
Takeaways & Limitations
The study uses binary severity classification rather than four categories: mild, common, severe, and critical.
Abstract
from arXiv · showhide
Background: Chest computed tomography (CT) is recognized as an important tool for COVID-19 severity assessment. As the number of affected patients increase rapidly, manual severity assessment becomes a labor-intensive task, and may lead to delayed treatment. Purpose: Using machine learning method to realize automatic severity assessment (non-severe or severe) of COVID-19 based on chest CT images, and to explore the severity-related features from the resulting assessment model. Materials and Method: Chest CT images of 176 patients (age 45.3$\pm$16.5 years, 96 male and 80 female) with confirmed COVID-19 are used, from which 63 quantitative features, e.g., the infection volume/ratio of the whole lung and the volume of ground-glass opacity (GGO) regions, are calculated. A random forest (RF) model is trained to assess the severity (non-severe or severe) based on quantitative features. Importance of each quantitative feature, which reflects the correlation to the severity of COVID-19, is calculated from the RF model. Results: Using three-fold cross validation, the RF model shows promising results, i.e., 0.933 of true positive rate, 0.745 of true negative rate, 0.875 of accuracy, and 0.91 of area under receiver operating characteristic curve (AUC). The resulting importance of quantitative features shows that the volume and its ratio (with respect to the whole lung volume) of ground glass opacity (GGO) regions are highly related to the severity of COVID-19, and the quantitative features calculated from the right lung are more related to the severity assessment than those of the left lung. Conclusion: The RF based model can achieve automatic severity assessment (non-severe or severe) of COVID-19 infection, and the performance is promising. Several quantitative features, which have the potential to reflect the severity of COVID-19, were revealed.
6 Department of Radiology Quality Control Center, Changsha, Hunan Province, 410008, China
COVID-19 severity assessment from chest CT is clinically important but difficult to perform manually at scale. This study develops a random-forest model using quantitative CT features to classify patients as non-severe or severe and identify severity-related features.
- Motivation: Manual chest CT severity assessment is labor-intensive and may delay patient isolation and treatment.
- Research gap: Few studies had addressed automatic COVID-19 severity assessment using chest CT images.
- Approach: The study builds a random forest model that uses quantitative chest CT features to classify COVID-19 severity as non-severe or severe.
- Approach: The model selects important quantitative features to reduce redundant and unrelated inputs before training.
Material and methods
The study constructs quantitative chest CT features and applies a random-forest framework for binary COVID-19 severity assessment. Features represent infection and attenuation-defined regions across whole, right, and left lungs and their subregions.
- Study population: 176 confirmed COVID-19 patients were included, with a mean age of 45.3±16.5 years and 96 male and 80 female participants.
- Clinical classification: Patients were grouped into non-severe and severe categories by combining mild/common and severe/critical clinical types.
- Feature extraction: A deep-learning CT analysis tool segmented lungs and infection regions, with 0.916 Dice similarity against manual delineations in 300 subjects.
- Feature extraction: Infection volume and ratio were calculated for the whole lung, right and left lungs, lobes, and segments.
- Feature extraction: Four HU ranges represented normal lung, GGO, consolidation, and vessel-calcification regions, whose volumes and ratios were also used as features.
- Feature set: The final feature set contained 63 quantitative features, including regional volumes and ratios and whole/right/left lung volumes.
- Modeling: A weighted random forest addressed the 11:5 non-severe-to-severe sample imbalance by assigning larger weights to minority-class samples.
Results
The random-forest model achieved promising binary COVID-19 severity-assessment performance, while feature analysis identified GGO measurements and right-lung features as particularly relevant.
- Model performance: 0.91 AUC was achieved by the K30 RF model on testing samples.The ROC curve was evaluated for the final severity-assessment model.
- Feature importance: GGO-region volume and its whole-lung volume ratio ranked among the most important quantitative features for severity assessment.The GGO region corresponded to the HU range [-700,-300].
- Feature importance: Right-lung quantitative features were relatively more important for severity assessment than corresponding left-lung features.The paper discusses this pattern in relation to the anatomical structure of the trachea and bronchi.
- Feature importance: GGO-region volumes and ratios were greater than consolidation-region volumes and ratios, with p=3.28 × 10^-7 in a pairwise t-test.The analysis used initial CT images from 176 patients.
- Scope and limitation: The study used binary classification because mild and critical COVID-19 cases were limited, combining mild and common as non-severe and severe and critical as severe.The authors state that future work will extend severity types using more patients through multi-center collaboration.