Downloaded from http://journals.lww.com/jcat by BhDMf5ePHKbH4TTImqenVBZZxeh5YHRL131uTbSEto7HnL0RxdbhoRXyAI8DSLZX on 10/18/2020 Computed Tomography Radiomics Can Predict Disease Severity and Outcome in Coronavirus Disease 2019 Pneumonia Fatemeh Homayounieh, MD,* Rosa Babaei, MD,† Hadi Karimi Mobin, MD,† Chiara D. Arru, MD,* Maedeh Sharifian, MD,† Iman Mohseni, MD,† Eric Zhang, MD,* Subba R. Digumarthy, MD,* and Mannudeep K. Kalra, MD* Purpose: This study aimed to assess if computed tomography (CT) radiomics can predict the severity and outcome of patients with coronavirus disease 2019 (COVID-19) pneumonia. Methods: This institutional ethical board–approved study included 92 patients (mean age, 59 ± 17 years; 57 men, 35 women) with positive re- verse transcription polymerase chain reaction assay for COVID-19 infec- tion who underwent noncontrast chest CT. Two radiologists evaluated all chest CT examinations and recorded opacity type, distribution, and extent of lobar involvement. Information on symptom duration before hospital ad- mission, the period of hospital admission, presence of comorbid conditions, laboratory data, and outcomes (recovery or death) was obtained from the medical records. The entire lung volume was segmented on thin-section Dig- ital Imaging and Communication in Medicine images to derive whole-lung radiomics. Data were analyzed using multiple logistic regression with re- ceiver operator characteristic area under the curve (AUC) as the output. Results: Computed tomography radiomics (AUC, 0.99) outperformed clinical variables (AUC, 0.89) for prediction of the extent of pulmonary opacities related to COVID-19 pneumonia. Type of pulmonary opacities could be predicted with CT radiomics (AUC, 0.77) but not with clinical or laboratory data (AUC, <0.56; P > 0.05). Prediction of patient outcome with radiomics (AUC, 0.85) improved to an AUC of 0.90 with the addition of clinical variables (patient age and duration of presenting symptoms be- fore admission). Among clinical variables, the combination of peripheral capillary oxygen saturation on hospital admission, duration of symptoms, platelet counts, and patient age provided an AUC of 0.81 for predicting pa- tient outcomes. Conclusions: Radiomics from noncontrast CT reliably predict disease severity (AUC, 0.99) and outcome (AUC, 0.85) in patients with COVID-19 pneumonia. Key Words: COVID-19, radiomics, chest CT, pneumonia, patient outcome (J Comput Assist Tomogr 2020;44: 640–646) KEY POINTS 1. Radiomics from unenhanced chest CT can accurately predict the extent of pulmonary involvement and patient outcome in COVD-19 pneumonia. 2. Adding clinical variables to radiomics resulted in modest im- provement in patient outcome prediction but did not improve the prediction of disease severity. 3. Clinical variables were inferior to radiomics for assessment of disease severity and patient outcome from COVID-19 pneumonia. Since its emergence in late 2019 in Wuhan, China, coronavi- rus disease 2019 (COVID-19) has affected more than 4.8 million people worldwide and has claimed more than 300,000 lives in 202 countries. 1 From the declaration of COVID-19 outbreak as a global health emergency in January 2020, the World Health Orga- nization upgraded it to a very high global risk in February 2020. By March 2020, public health crises from COVID-19 pneumonia had reached pandemic proportions throughout the world. Coronavirus disease 2019 pneumonia results from an extremely infectious β-coronavirus that targets the angiotensin-converting enzyme II and inflicts direct lung injury. Among vulnerable subjects, particularly the elderly and patients with comorbid diseases such as chronic obstructive pulmonary diseases, diabetes mellitus, and car- diovascular diseases, the resultant lung injury from COVID-19 pneumonia can progress rapidly to diffuse alveolar damage and acute lung failure. 2,3 Given the highly contagious nature of the in- fection, the burden of COVID-19 pneumonia has imposed substan- tial constraints on most health care systems from the underdeveloped nations to the most developed countries in the world. Diagnosis of COVID-19 infection is made with reverse tran- scription polymerase chain reaction (RT-PCR) assays with detec- tion of specific nucleic acid of severe acute respiratory syndrome coronavirus 2 in oral or nasopharyngeal swabs. Chest computed to- mography (CT), although initially proposed as more sensitive than RT-PCR assay for diagnosis of COVID-19 infection, 4,5 was later reported as negative in more than 20% of patients. 6 The American College of Radiology and the Society of Thoracic Radiology have both recommended against the use of chest CT for diagnosis or screening of patients with COVID-19 pneumonia. 7,8 Chest CT is frequently used to assess the severity and complications of COVID-19 pneumonia. Several studies have described that chest CT examinations are useful for evaluating disease stage and severity. 9–17 Most studies have focused on qualitative assessment and grading of pulmonary involvement in each lung lobe to estab- lish disease severity, which can be both time-consuming and asso- ciated with interobserver variations. 11–17 METHODS Approvals and Disclosures The institutional ethical board approved our retrospective study with the waiver of informed consent. We have no financial disclosures pertaining to this article. Our institution received From the *Department of Radiology, Massachusetts General Hospital and the Harvard Medical School, Boston, MA; and †Department of Radiology, Firoozgar Hospital and Iran University of Medical Sciences, Tehran, Iran. Received for publication May 20, 2020; accepted July 13, 2020. Correspondence to: Fatemeh Homayounieh, MD, Department of Radiology, Massachusetts General Hospital, 75 Blossom Court, Room 248, Boston, MA 02114 (e‐mail: fhomayounieh@mgh.harvard.edu). The authors have not received any funding for this study and have no financial disclosures pertaining to this article. Thir institution received research grants from Siemens Healthineers, Lunit Inc, and Riverain Tech. for unrelated projects. The institutional ethical board approved this retrospective study. Copyright © 2020 Wolters Kluwer Health, Inc. All rights reserved. DOI: 10.1097/RCT.0000000000001094 ORIGINAL ARTICLE 640 www.jcat.org J Comput Assist Tomogr • Volume 44, Number 5, September/October 2020 Copyright © 2020 Wolters Kluwer Health, Inc. All rights reserved.