COVITALE 2020 from eastern Indian population: imageologists perspective, a learning curve

Background High-resolution computed tomography (HRCT) chest becomes a valuable diagnostic tool for identifying patients infected with Coronavirus Disease 2019 (COVID-19) in the early stage, where patients may be asymptomatic or with non-specific pulmonary symptoms. An early diagnosis of COVID-19 is of utmost importance, so that patients can be isolated and treated in time, eventually preventing spread of the disease, improving the prognosis and reducing the mortality. In this paper, we have highlighted our radiological experience of dealing with the pandemic crisis of 2020 through the study of HRCT thorax, lung ultrasonography, chest X-rays and artificial intelligence (AI). Results Results of CT thorax analysis have been given in detail. We had also compared CT severity score (CTSS) with clinical and laboratory parameters. Correlation of CTSS with SpO2 values and comorbidities was also studied. We also analysed manual CTSS with the CTSS scored calculated by the AI software. Conclusions CTSS and use of COVID-19 Reporting and Data System (CORADS) result in accuracy and uniform percolation of information among the clinicians. Bed-side X-rays and ultrasonography have played a role where the patients could not be shifted for CT scan. The possibility of predicting impending or progression of hypoxia was not possible when SpO2 mapping was correlated with the CTSS. AI was alternatively tried with available software (CT pneumonia analysis) which was not so appropriate considering the imaging patterns in the bulk of atypical category.


Background
A febrile lower respiratory infection of unknown aetiology was reported in Wuhan, China, in December 2019 and was later discovered to be caused by a novel coronavirus, a severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2) [1]. Since then, it has spread fast over the world, infecting millions of people [2].
Because it has already been shown that high-resolution computed tomography (HRCT) chest has excellent accuracy for early diagnosis, triage, proper patient management and follow-up, radiology has become a pioneer branch for identifying individuals infected with Coronavirus Disease 2019   [3,4]. Artificial intelligence (AI)-assisted imaging acquisition may substantially assist in automating the scanning approach and reshaping the workflow with little patient interaction, ensuring the greatest protection for imaging specialists. In addition, AI can boost productivity by accurately delineating infections in X-ray and CT images, allowing for more efficient quantification [5].
Since most lung parenchymal lesions in COVID-19 are distributed peripherally, these lesions may be visualized on lung ultrasonography (LUS) in most of the cases [6,7].
The main aim of this article is to share overall radiological experience in management of COVID-19. We looked at the under-appreciated modality, lung ultrasonography and the overhyped artificial intelligence in COVID patients, in addition to HRCT of thorax and X-rays.

Study design
This prospective observational study was done at Odisha COVID Hospital, KIMS, conducted over a period of 10 months between March and December 2020. Ethical committee clearance has been taken for the research purpose. A study of 4338 COVID-19 patients who had undergone HRCT chest was scrutinized by experienced radiologists, blindfolded to any other associated findings. Out of these 4338 patients, 2000 images were also scrutinized with AI software. Seriously ill-intensive care unit (ICU) patients (124 patients) were studied by lung ultrasound and X-rays helped in 1402 patients.
The conventional gold standard was real-time reverse transcription polymerase chain reaction (rRT-PCR) test, for the qualitative detection of nucleic acid from SARS-CoV-2 in upper and lower respiratory specimens (such as nasopharyngeal or oropharyngeal swabs, sputum, lower respiratory tract aspirates or bronchoalveolar lavage).

Inclusion criteria
For HRCT thorax, the inclusion criterion was laboratoryproven RT-PCR-positive cases. Only critically sick ICU patients who were unable to undergo HRCT were considered for LUS.

Exclusion criteria
Clinically suspected COVID-19 cases that were negative on RT-PCR, as well as COVID-19-positive patients who had not underwent HRCT chest, were eliminated from the research. Pregnant females and patients with severe artefacts on CT images were also excluded.

Epidemiological and clinical data collection
All relevant epidemiological data were collected from hospital records. Based on the clinical data, the patients were divided into asymptomatic and symptomatic groups. The patients were then categorized into three groups (mild, moderate and severe), based on the clinical severity (Table 1).

CT image data acquisition
Examination of the patients with HRCT chest was performed with dedicated 64 Slice CT scanner (Siemens, Somatom: Go up). Scanning parameters were 1 mm slice thickness, 5 mm gap, 120 kV and 150 mA. The scan was done generally within 2 days of hospitalization.

Image interpretation and analysis
The HRCT chest images were viewed with standard lung (window width 1500 HU; window level -600 HU) and mediastinal (window width 350 HU; window level 40 HU) settings.
Semi-quantitative assessment for extent of lung parenchymal involvement in CT chest images was done using CT severity score (CTSS). Each lobe of the lungs (total five lobes) was visually assessed and given a score between 0 and 5 on the basis of percentage of parenchymal involvement: 1 for < 5% involvement, 2 for 5-25% involvement, 3 for 26-49% involvement, 4 for 50-75% involvement and 5 for > 75% involvements. The CTSS is the sum of all lobar scores and can range from 0 (no involvement) to 25 (maximum involvement).
Changes in CT findings with duration of symptoms were assessed by classifying the study population into three phases (early, intermediate and late phase), based on the duration between day of onset of illness (day of onset of symptoms in symptomatic patients or day of becoming positive on RT-PCR in asymptomatic patients) and day of CT scan. Patients were considered to be in the (a) early phase of illness if the duration was < 5 days, (b) intermediate phase of illness for 6-10 days duration and (c) late phase if the scan was done > 10 days after the date of onset of illness.
"Choicemmed MD300C2 Pulse Oximeter" was used to measure patient's SpO2 level at the time of HRCT thorax and again after 5 days.

Automated processing of chest CT
Artificial intelligence-based algorithm "CT pneumonia analysis" based on 3D deep learning developed by Siemens Healthineers was used in our study. This algorithm identifies and quantifies the hyperdense areas (areas of high opacity) of lungs. It performs automated lung opacity analysis on axial CT data with slice Table 1 Clinical severity score according to Ministry of Health and Family Welfare, Government of India thicknesses up to 5 mm, resulting in multiplanar reformatting (MPR) series containing segmentations of the high opacity abnormalities of the lungs and provides a table with various measurements, e.g. the relative ("percentage of opacities") and absolute volume of the lungs affected by opacities. Additionally, the mean and standard deviation of Hounsfield unit (HU) values between lung parenchyma and the detected opacities can be compared. Segmentation was done using multi-scale deep reinforcement learning and deep image-to-image network (DI2IN) [9,10]. The algorithm quantifies the commonly seen CT abnormalities (i.e. GGO and consolidation) and categorized the data into two combined severity measures: (a) lung severity score (LSS), lung high opacity score (LHOS) and (b) percentage of opacity (PO), percentage of high opacity (PHO) [11]. Lung severity score, which measures the extent of lung involvement, is the sum of the individual lobe severity scores. All voxels within the lung containing either full or partial GGO or consolidations are considered as positive and given a single label and rest of images within and outside of lungs are defined as negative.
The Pearson's correlation coefficient (r) was used to demonstrate linear relationship between predicted and manual (actual) values. The degree of correlation was assessed using interclass correlation (ICC) coefficient based on single rater, consistency and two-way random effect model. In order to further assess the level of agreement between two methods of LSS, Bland-Altman analysis was used. ANOVA test was used to find the difference in the categorical data between the three clinical severity groups and a two-tailed p value was calculated. p Value between the categorical data less than 0.05 indicated statistical significant difference.

Lung ultrasonography (LUS) examination
Lung ultrasound was performed with a convex (2-6-MHz) transducer set at depth of 6.8 cm and linear transducer (10-15-MHz) set at depth of 11.6 cm. The LUS examination was performed by two radiology residents both having experience of 2.5 years. The intraclass correlation coefficient (ICC) was determined. An almost perfect agreement for the Lung segmental scores (LUSS) was noted (ICC 0.82, 95% CI 0.57-0.95).
The thorax was scanned in 12 lung areas [12][13][14]. The COVID-19 LUS in emergency protocol (CLUE) [15] was used. The LUSS points ranged from (normal) 0 to 3, with higher points allocated to severe lung changes. Based on the total score from 12 lung zones, the severity was classified as mild (score 1-5), moderate (> 5-15) and severe (> 15). Follow-up LUS examinations were performed in all patients after 7 (mean) days of the initial LUS examination. The LUS images were electronically stored and analysed.
The ICC was used to assess the degree of agreement between LUSS and CT segmental scoring (CT SS). An ICC < 0.50 was considered poor that from 0.50 to 0.75 moderate, that from 0.75 to 0.90 good and that from 0.90 to 1 excellent. Mean values were reported along with 95% CIs. Statistical significance was set at p < 0.05 using McNemar's Chi-square test. Cohen's kappa (k) test was used to compare abnormal chest CT findings with abnormal LUS findings using the scoring systems.

X-ray examination
All chest X-rays (CXR) were acquired with portable X-ray units (100 mA Siemens Multimobile 2.5) and processed with Care stream DV 6950 CR system. CXR were obtained in the postero-anterior (PA) or antero-posterior (AP) projection as per the condition of patient.

Results
Serial data from 4338 COVID-19-positive patients, qualifying the inclusion criteria, were compiled and analysed ( Fig. 1 All images were reviewed independently by three radiologists (5, 5 and 30 years of experience). Using intraclass correlation coefficients, comparisons between observers for HRCT thorax analysis showed acceptable interobserver reliability (ICC 0.82, 95% CI 0.74-0.89). Average CTSS in this study was found to be 4.9. Among asymptomatic patients, 86.6% cases had CTSS of zero and 13.4% had scores more than zero. 91.3% of cases having score ≤ 7 were asymptomatic or mild. On the other hand, 75.1% cases having CTSS of 8-14 were clinically moderate and 81.2% cases having score ≥ 15 were clinically severe (Table 4). So, the positive predictive values (PPVs) of CTSS ≥ 15 for detecting clinically severe and symptomatic patients are 81.2% and 98.5%, respectively. The association between CTSS grading and clinical severity was found to be statistically significant (Chi-square value = 2654, df = 9, p < 0.001).

Temporal changes in CT findings with duration of symptoms
Among the radiologically positive cases, 1492 (51.9%) patients were in early phase (≤ 7 days), 860 patients (29.9%) in intermediate phase (8- The presence of pulmonary nodules in images of some COVID patients led to the suspicion of associated tuberculosis, which is an endemic in this part of the subcontinent. Atypical category of COVID 19 couldn't be differentiated from tuberculosis in HRCT chest because of similar imaging features (Fig. 5). Accurate diagnostic tests are essential for both TB and COVID-19. Since sputum could not be collected due to the constraints of fear psychosis and patients' condition, testing with IGRA for exclusion of TB was resorted to. We conducted a study in 50 COVID patients over a period of 6 months suspicious of tuberculosis. Interferon-gamma release assay (IGRA) test was undertaken. IGRA test was found to be negative in 42 patients and positive in 8 patients. We concluded that it is not always possible to differentiate features of atypical COVID-19 from tuberculosis on the basis of HRCT chest alone. Hence, exclusion of tuberculosis will need supportive relevant laboratory investigations (sputum AFB, CB NAAT and IGRA) for appropriate diagnosis and management.

Artificial intelligence analysis
Percentage of opacity (PO) and percentage of high opacity (PHO) in HRCT chest were obtained using "CT pneumonia analysis algorithm" (Fig. 6). The linear correlation by calculating Pearson's correlation coefficient (r) value was 0.88 which suggests strong linear correlation between the LSS of AI with actual values (Fig. 7). Intraclass correlation coefficient (ICC) and their 95% confidence intervals were calculated to be 0.87, 0.86 and 0.88, respectively. The ICC of 0.8 shows good reliability between actual LSS and LSS assessed by AI-based algorithm. The confidence interval shows 95% of the times ICC will lie between 0.86 and 0.88. Bland-Altman plot shows good agreement between two methods (Fig. 8).

CT staging with duration of hospital stay and mortality
In our study, 340 (7.8%) patients died during an average follow-up of 14.2 ± 4 days (range 1-24 days). Mortality was higher in patients ≥ 60 years of age (53.3%) than in patients < 60 years of age (46.7%) (p < 0.003), with a mean mortality age of 51 years. The average duration of stay in hospital was found to be 12 days (range 9 to 24 days). Patients with CTSS of < 5 had less mean duration of stay (10 days) and did not require any active management. Patients with CTSS of 5 to 15 had mean duration of 13 days; most of them recovered after proper management. Patients with CTSS > 15 were severely ill patients with relatively higher mortality of 32.9%. These patients were kept in ICU. Correlation of CTSS with laboratory markers was undertaken (Table 5). p Value was calculated, and Pearsons' correlation method was used to find the correlation coefficient (r).
The trend of severity (on the basis of CTSS) in eastern Indian population over the periods of month (from March to December) is shown in (Fig. 9).

USG thorax in COVID
We studied LUS in 124 ICU patients. The presence of B lines was the most common finding in 118 (95.16%) of the cases. Pleural line thickening in > 6 lung areas was seen in 98/124 (79.03%). The right lung was affected in 100/124 (80.5%) and the left lung in 96/124 (78.7%). LUS imaging features are shown in Fig. 10.
Confluent B lines originating from regular pleural lines which were previously described as "waterfall" artefacts ( Fig. 10b) and were supposed to represent an early stage of actively spreading COVID-19 pneumonia alternating with areas of normal lung parenchyma were observed in 76/124 (61.2%) of the cases.
An interesting finding also encountered in this study was the narrow and bright B line originating from the thickened pleura which was brighter and narrower than the regular separate B lines and did not tend to fan out till the end of the screen. We are uncertain if this is a variety of the already described "waterfall" and "light beam sign" which originate from a larger portion of the regular pleural line. We felt this was similar to a "torchlight" a "single  light beam" being flashed in a sense that it was extremely bright undoubtedly brighter than the Z line and B line, was erased by the A lines and almost always was from a thickened pleura. These "torchlight" B lines (Fig. 10c) were observed in 32 (25.8%) of the patients and were seen more in the antero-superior and antero-inferior areas of bilateral lungs typically in clinically severe cases. Further validation will be needed to establish if it can be an indicator of severe COVID pneumonia.

Discussion
HRCT chest findings were analysed and correlated with symptomatology and severity of patients to explore possible roles of HRCT in diagnostic, therapeutic and prognosis of COVID-19 disease. A total of 4338 images were studied. Out of asymptomatic patients, 39.8% were found to be radiologically positive. This significant percentage of asymptomatic COVID-19 patients pose a real threat to society as they may act as silent carriers for transmission of virus. The prominent HRCT chest features were bilateral ground-glass opacities (GGOs) with or without  consolidation. In this study, nearly two-third patients had positive CT findings with most patients (55.8%) showing a mixed pattern of GGO and consolidation. The distribution of the lesions in our study was mostly peripheral and lower lobar with multilobar and bilateral lung involvement seen in 66% patients. The predilection for these areas has also been reported previously [16]. Thin or thick subpleural band-like opacities usually in the basal segments of lower lobes and some specific patterns like interstitial/septal thickening (crazy paving/reticular pattern), reverse halo sign and halo sign were seen in a significant proportion of patients. These findings corroborate with the findings of the study done by Ye et al. [17]. Positive HRCT chest findings were also seen in some asymptomatic patients, especially the young, may be due to the differences in disease tolerability and immune status. 91.3% of cases having score ≤ 7 were asymptomatic or mild. On the other hand, 75.1% cases having CT severity score 8-14 were clinically moderate and 81.2% cases having score ≥ 15 were clinically severe. The CT severity scores (CTSS) showed good correlation with clinical status of patients, as seen in previous studies [18]. It was also observed that clinically severe patients showed higher CTSS compared to the non-severe patients. The positive predictive values (PPVs) of CTSS ≥ 15 for detecting clinically severe and symptomatic patients are 81.2% and 98.5%, respectively.
These data suggested that CTSS may be an informative indicator to predict the severity of the disease. The proposed grades of CTSS are (1) mild (CTSS: 0-7), (2) moderate (CTSS: [8][9][10][11][12][13][14] and (3) severe (CTSS: [15][16][17][18][19][20][21][22][23][24][25]. This grading can help to prognosticate and tailor the clinical management of patients. HRCT chest scans of patients in the early phase (0-7 days) of disease show mostly GGO and in late stage (> 14 days) consolidation tends to be a predominant feature. In the early stages, single or multiple small GGO, consolidation and mild interstitial thickening could be seen. In the intermediate phase (8-14 days), the number of involved lobes was more with increase in consolidation and septal thickening in the involved lobes which correlated with increased disease severity. The diffuse lesions involving both peripheral and central parts of lungs, causing "white lungs, " appearance were seen in the most severely affected patients.
The other observation in our study was that, even if the total area of lung involved remained the same, features   [19]. Computed tomography chest imaging also showed some non-specific findings which include pleural effusion, pulmonary nodules and thoracic lymphadenopathy. Majority of cases with pleural effusion were in the intermediate phase of the disease. On LUS, confluent B lines originating from regular pleural lines which were supposed to represent an early stage of actively spreading COVID-19 pneumonia were observed in 76/124 (61.2%) of the cases; however, other studies which were not in the ICU setting reported a higher incidence of confluent B lines [13,14,20]. In our study, the subpleural kind of consolidation was more common than the "starry sky" pattern. This was a deviation from the previous studies which reported a higher incidence of the "starry sky" pattern usually involving more of the lung parenchyma [20,21].
It was reported that approximately 50 percent of patients were confirmed to have elevated D-dimer levels and that high D-dimer rates correlated with poor prognosis [28,29]. Moderate positive correlation (r = 0.56) with a p value of < 0.001 remained as supporting biomarker in our series. Pearson's correlation of C-reactive protein (CRP) levels in 1858 patients indicated p value of < 0.0001, i.e. moderately positive and useful predictor of illness. Levels of CRP were found to be increased in 74% cases with mean CRP levels of 78 mg/L (Range 24 -223 mg/L) among the total deaths. Procalcitonin is now being seen to be an important indicator for early detection low risk of bacterial coinfection in the ongoing COVID-19 pandemic [28]. Our observation also suggests mild positive correlation of procalcitonin and ESR in determining the severity of COVID, but since the correlation is mild, both cannot be considered as reliable markers of the severity. Considerable proportion of patients showed lymphopenia. Associated and iatrogenic complications of the disease like pneumomediastinum and pneumothorax were well picked up on HRCT for early management.
Manual assessment of chest CT in COVID-19 pandemic scenario was relatively more stressful and timeconsuming. Hence, there was a need for rapid, precise and automated method of detection and quantification of common findings of COVID-19 in chest CT. The "CT Pneumonia Analysis" AI algorithm gives multiple quantitative parameters which are helpful in assessing the extent of individual lobe involvement as well as combined whole lung involvement. The Pearson's correlation coefficient showed strong linear relationship between both methods in assigning LSS. Intraclass correlation coefficient showed good reliability and Bland-Altman plot showed good agreement between AI LSS and actual LSS reported manually. However, the software ("CT Pneumonia Analysis") could not detect GGOs in early stage of disease and also other atypical findings like pleural and pericardial effusions, mediastinal lymphadenopathy, nodules and complications like pneumothorax, pneumomediastinum, surgical emphysema in contrary to conventional method. It needs appropriate modifications and training for detecting these additional features.
There was a trend of increasing severity from month of July onwards as compared to months of April to June. In April-May the proportion of patient having CTSS 16-25 was 4%, which touched 27% in August and September. It slightly decreased to 24% during last quarter. The sudden change in trend of severity could be attributed to the migrants labourers, domestic and international travellers, who came back post lockdown. Our assumption was that this category of patients could have been infected by more virulent viral strain.

1.
We examined all the main radiological modalities, with the exception of lung MRI, which has a debatable usefulness, especially in COVID-19. 2. The AI software which we used in our study was unable to recognize atypical features of COVID-19.

Conclusions
COVID-19 could be well classified as typical, atypical and indeterminate categories. Diabetics, aged population, lower socioeconomic strata, males were more vulnerable.