One hundred and twenty patients who were tested positive for novel coronavirus by nasopharyngeal swap were enrolled in our retrospective study in the period of 1 May and 20 June 2020. There were 98 males and 22 females with a male to female distribution of 4.5:1 and with an age range from 28 to 83 years old with mean = 52.63 ± 12.79.
The study protocol was approved by the local Ethics Committee. All patients provided a written informed consent.
Patients were stratified into three clinical groups based the WHO interim guidance [20, 21]: group A, mild cases; group B, severe cases; and group C, critical cases. Group A involves patients with mild clinical symptoms in the form of fever, mild respiratory tract manifestations, and positive CT findings of pneumonia. Group B involves patients with respiratory rate ≥ 30 times per minute, oxygen saturation ≤ 93% at rest, arterial oxygen partial pressure (PaO2)/inspired oxygen (FiO2) ≤ 300 mmHg (1 mmHg = 0.133 kPa), or significant progression of pneumonia CT findings within 24–48 h ≥ 50%. Group C involves patients that are admitted to the intensive care unit for mechanical ventilation or had a FiO2 of at least 60% or more.
Image acquisition and analysis
All CT examination was performed using two multidetector CT scanners (Somatom Perspective, Siemens, Germany, and Optima CT 540, GE, America), using the following parameters: tube voltage = 120 kVp, tube current (regulated by automatic dose modulation), 30–75 mAs, pitch = 1–1.25 mm, matrix = 512 × 512, slice thickness = 5 mm, and FOV = 350 mm × 350 mm.
Image reconstruction was done at a slice thickness of 1–1.25 mm. All were the initial CT scans at the time of patients’ admission and are performed as non-contrast studies. Two experienced radiologists (20 years of experience) independently reviewed all the scans, and they were blinded to the patients’ clinical and laboratory data.
Qualitative image analysis
CT severity score was estimated for each one of the five lung lobes by calculating the dissemination of the chest manifestations (opacity), namely the ground-glass opacities (GGO), consolidation, crazy-paving pattern, septal thickening, and pulmonary fibrosis giving score (0–4) for 0, 25, 50, and ≥ 75% involvement, respectively, with the sum representing the total severity scores for the whole lung (0–20).
Previous studies [3, 4] reported that the degree of consolidation and crazy-paving pattern was highly suggestive for the disease progression/peak, so we used a total sum extent of crazy-paving and consolidation as an indicator for the disease severity. The severity score for the consolidation and crazy-paving was calculated for each lobe using the same criteria (0–4 scores), and the total score for the lungs is the sum of individual lobes (0–20 scores).
Quantitative image analysis
CT Pneumonia Analysis algorithm is designed by Siemens Healthineers to automatically identify and quantify abnormal tomographic patterns in the lungs from chest CT for research purposes. The system takes as input a non-contrasted chest CT, and identifies and 3D segments the lungs and lobes before segmenting the abnormalities. It outputs two combined measures of the severity of lung/lobe involvement, quantifying both the extent of COVID-19 abnormalities and presence of high opacities. High opacity abnormalities were shown to correlate with severe symptoms. The first disease severity measure is global, while the second is lobe-wise:
-
First global measure
-
Percentage of opacity (PO): percentage of predicted volume of abnormalities compared to the total lung volume
-
Percentage of high opacity (PHO): percentage of predicted high opacity volume compared to the predicted volume of abnormalities
-
Second lobe-wise measure
The computed results could be used to analyze the severity and monitor the progression of abnormalities in patients exhibiting COVID-19 symptoms.
AI-Rad Companion Research CT Pneumonia Analysis
The family of AI-powered augmented workflow solutions, running on the teamplay digital health platform, helps to reduce the burden of basic repetitive tasks and increase the diagnostic precision when interpreting medical images. Its solutions provide automatic post-processing of imaging datasets through AI-powered algorithms. The automation of routine workflows with repetitive tasks and high case volumes helps to ease the daily workflow, so that the radiologist can focus on more critical issues. This system is capable of computing the severity scores in approximately 10 s per case versus 30 min for manual annotations. These results could be used to rapidly assess the extent of lung infection and monitor the progression of abnormalities in patients exhibiting COVID-19 symptoms.
Using an artificial intelligence algorithm, the abnormal tomographic patterns commonly present in lung infections, namely ground-glass opacities (GGO) and consolidations, were automatically detected and quantified. This algorithm estimates the overall lung affection and quantifies the high opacity abnormalities using a 3D segmentation of lesions, lungs, and lobes.
Opacity score is calculated for each lobe by estimating the given region percent opacity as follows: score= 0, ≤ 25%; score = 1, 25–50%; score = 2, 50–75%; score = 3, > 75%; and score = 4 and the total score is the sum of these values.
Variable parameters are also obtained including lung volume (ml), volume of opacity (ml), percentage of opacity within a given lung region (%), volume of high opacities as absolute value (ml), a given lung region percentage of high opacities, total mean HU, given lung region mean HU of opacity, total HU standard deviation, and a given lung region opacity HU standard deviation. All these parameters are calculated for the whole lung, left lung, right lung, and per lung lobe, respectively.
Statistical analysis of the collected data
Data were expressed in number (no.), percentage (%), mean (\( \overline{x} \)), and standard deviation (SD) and statistically analyzed by an IBM-compatible personal computer with SPSS statistical package version 23 (SPSS Inc. Released 2015. IBM SPSS statistics for windows, version 23.0, Armonk, NY: IBM Corp.).
ANOVA test was used for the comparison of quantitative variables between more than two groups of normally distributed data with Tukey’s test as the post hoc test while the Kruskal-Wallis test was used for the comparison of quantitative variables between more than two groups of not normally distributed data with Tamhane’s test as the post hoc test.
Pearson’s correlation was used to show correlation between two continuous normally distributed variables while Spearman’s correlation was used for not normally distributed ones.
-
Chi-square test (χ2) was used to study association between qualitative variables. Whenever any of the expected cells were less than five, Fisher’s exact test was used. Z test was used to compare column proportions.
-
Receiver operating characteristic (ROC) with respective points of maximal accuracy for sensitivity and specificity was generated to determine radiological variables’ performance. Area under the ROC curve (AUROC) measures the accuracy of the test. An area of 1 represents a perfect test; an area of 0.5 represents a worthless test. Two-sided P value of < 0.05 was considered statistically significant.