This study was approved by the ethics committee of Zhongda Hospital (2020ZDSYLL013-P01 and 2020ZDSYLL019-P01), and informed consent was waived due to the emergent event of the pandemic.
Clinical information and imaging data for all COVID-19 patients diagnosed in Jiangsu before February 18, 2020, were obtained, retrospectively from electronic medical records, established by the Department of Health, Jiangsu Province.
Patients with no clinical records or with missing, incomplete, or poor-quality chest CT images were excluded. Figure 1 shows the detailed flowchart of this study.
Based on the 5th edition of the guideline for the diagnosis and treatment of COVID-19, published by the National Health Commission of the People’s Republic of China on February 8, 2020, patients were assigned into three groups, classified according to disease severity: asymptomatic/mild, moderate, or severe/critically ill. Patients were classified as (i) “asymptomatic” if they had no symptoms of disease, (ii) “mild” if they had mild clinical symptoms, but no imaging abnormality, and (iii) “moderate” if they had one or more symptoms (fever, cough, diarrhea, etc.), and imaging showed manifestations of pneumonia or (iv) “severe” if they had one of the following conditions: (a) respiratory distress presenting with respiratory rate ≥ 30 beats/min, (b) mean oxygen saturation in resting state ≤ 93%, (c) arterial blood oxygen partial pressure/oxygen concentration ≤ 300 mmHg. (v) Patients with either shock, respiratory failure requiring mechanical ventilation, or combined organ failure requiring admission to an intensive care unit (ICU) were classified as “critically ill.”
In order to analyze the evolution of imaging and clinical characteristics, the day when the initial symptoms emerged was defined as day 1. When no initial symptoms were reported, the date of outpatient visit was used as day 1. The time points for each CT scan, change of severity grading relative to day 1, were recorded and were further assigned to days 1–3, days 4–6, days 7–9, days 10–12, days 13–15, and beyond day 15.
CT imaging was performed using multislicer CT scanners. Thin-section images were collected preferentially (for further details relating to the acquisition of CT images, please see Table S1). All raw data with format of Digital Imaging and Communications in Medicine were then transferred to work stations for post-processing.
Lung segmentation, lesion extraction, and labeling were performed using a dedicated artificial intelligence (AI) system with deep learning algorithm for pulmonary pneumonia (Deepwise & League of PhD Technology Co., Ltd.). The AI algorithm achieved a dice of 0.97 in the HOUSE dataset (easy) and 0.89 in the LOLA dataset (hard), which proved its robustness. The accuracy of the AI system in lesion extraction and labeling was also manually checked by an investigator (Y.W.) with more than 10 years of experience in chest imaging. Bilateral lungs were segmented into the right upper lobe, right middle lobe, right lower lobe, left upper lobe, and left lower lobe automatically, and extraction of pulmonary opacities was carried out at the same time. All extracted opacity pieces were annotated as GGO or consolidation by the AI system and were approved by an investigator (Y.W.). The masks of opacities and each lung’s lobes were saved for further voxel-based measurements including volume, X-ray attenuation, and location. Detailed information regarding segmentation and lesion extraction are available in supplemental methods.
Further assessment of segmented lobes and extracted opacities was performed as follows:
Volume measurement: it included volume of bilateral lungs, GGO, consolidation, GGO + consolidation, and aerated lung.
Density measurement: it included the X-ray attenuation of bilateral lungs, overall opacities, opacities in upper lobes including bilateral upper lobes and right middle lobe, and opacities in bilateral lower lobes.
Location analysis: A standard lung was selected from a healthy 37-year-old male without lung abnormalities. Subsequent registration and projection of all included chest CT images to the standard lung was performed. Voxel-based frequencies of opacity were measured by drawing five volumes of interest (VOIs) in the standard lung, with each of the VOI larger than 15 mL. Four VOIs were placed in the subpleural areas (within 15 mm from the pleura), which were the right posteroinferior subpleural area, right posterosuperior subpleural area, right anterior subpleural area, and right medial subpleural area, and the last one was placed in the right central area 20 mm away from the pleura (Fig. S1). The number of lobes with pulmonary opacities was also recorded and scored from 0 to 5, with 5 indicating all 5 lobes were involved.
Dynamic evolution: Dynamic changes of volume, density, and frequency of opacities by location from symptom onset (day 1) to beyond day 15 were investigated.
Clinical and laboratory characteristics
Demographic information, exposure type, initial symptoms (fever, cough, sputum, shortness of breath, dyspnea, anorexia, or diarrhea), possible exposure time, time of initial symptoms, date of admission to hospital, comorbidity, and clinical laboratory findings including arterial oxygen saturation (SaO2), blood cell counts (blood leukocyte count, lymphocyte count, and platelet count), and biomarkers of inflammation (C-reactive protein level, procalcitonin level, lactate dehydrogenase), hepatic and renal function (aspartate aminotransferase level, alanine aminotransferase level, creatinine level), and coagulation (D-dimer level) were collected.
All consecutive data were listed as means ± standard deviations (SDs) or medians with interquartile ranges (IQRs) for Gaussian and skewed distributed data. Skewed distributed data were tried to convert to Gaussian distribution by logarithmic transformation for variance analysis. The Chi-square test, Kruskal-Wallis test, or Mann-Whitney test was used to examine statistical differences for ratio and skewed distributed data. All statistical analyses were performed using the R statistical software version 3.0.3.