[Paper Review] Severity Assessment of Coronavirus Disease 2019 (COVID-19) Using Quantitative Features from Chest CT Images

Zhenyu Tang, Wei Zhao+5 more|arXiv (Cornell University)|Mar 26, 2020

COVID-19 diagnosis using AIMedicine19 references149 citations

TL;DR

The paper develops a random forest model to automatically classify COVID-19 severity (non-severe vs severe) from chest CT quantitative features, achieving strong AUC and accuracy with a 3-fold cross-validation on 176 patients.

ABSTRACT

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.

Motivation & Objective

• Motivate rapid, automatic severity assessment to assist timely treatment for COVID-19 patients.
• Investigate quantitative CT-derived features that reflect disease severity.
• Build and evaluate a machine learning model to distinguish non-severe from severe cases based on imaging features.

Proposed method

• Extract 63 quantitative CT features including whole-lung infection volume and ground-glass opacity (GGO) metrics.
• Train a random forest classifier to predict severity (non-severe vs severe).
• Compute feature importance from the RF model to reveal severity-related features.
• Evaluate using three-fold cross-validation to report performance metrics.

Experimental results

Research questions

• RQ1 Can quantitative CT-derived features discriminate between non-severe and severe COVID-19?
• RQ2 Which CT features (and lung-side contributions) are most predictive of severity?
• RQ3 How well does a random forest model perform in automatic severity assessment on the provided dataset?

Key findings

• RF model achieves 0.933 true positive rate (TPR).
• RF model achieves 0.745 true negative rate (TNR).
• RF model achieves 0.875 overall accuracy.
• RF model achieves 0.91 area under the ROC curve (AUC).
• Volume and volume ratio of ground-glass opacity (GGO) regions strongly relate to severity.
• Features from the right lung show stronger relation to severity than those from the left lung.