Convolutional Neural Network-Based Automatic Analysis of Chest Radiographs for the Detection of COVID-19 Pneumonia: A Prioritizing Tool in the Emergency Department, Phase I Study and Preliminary "Real Life" Results.

The aim of our study is the development of an automatic tool for the prioritization of COVID-19 diagnostic workflow in the emergency department by analyzing chest X-rays (CXRs). The Convolutional Neural Network (CNN)-based method we propose has been tested retrospectively on a single-center set of 542 CXRs evaluated by experienced radiologists. The SARS-CoV-2 positive dataset (n = 234) consists of CXRs collected between March and April 2020, with the COVID-19 infection being confirmed by an RT-PCR test within 24 h. The SARS-CoV-2 negative dataset (n = 308) includes CXRs from 2019, therefore prior to the pandemic. For each image, the CNN computes COVID-19 risk indicators, identifying COVID-19 cases and prioritizing the urgent ones. After installing the software into the hospital RIS, a preliminary comparison between local daily COVID-19 cases and predicted risk indicators for 2918 CXRs in the same period was performed. Significant improvements were obtained for both prioritization and identification using the proposed method. Mean Average Precision (MAP) increased (p < 1.21 × 10−21 from 43.79% with random sorting to 71.75% with our method. CNN sensitivity was 78.23%, higher than radiologists' 61.1%; specificity was 64.20%. In the real-life setting, this method had a correlation of 0.873. The proposed CNN-based system effectively prioritizes CXRs according to COVID-19 risk in an experimental setting; preliminary real-life results revealed high concordance with local pandemic incidence.

Emergency room comprehensive assessment of demographic, radiological, laboratory and clinical data of patients with COVID-19: determination of its prognostic value for in-hospital mortality.

Mortality risk in COVID-19 patients is determined by several factors. The aim of our study was to adopt an integrated approach based on clinical, laboratory and chest x-ray (CXR) findings collected at the patient's admission to Emergency Room (ER) to identify prognostic factors. Retrospective study on 346 consecutive patients admitted to the ER of two North-Western Italy hospitals between March 9 and April 10, 2020 with clinical suspicion of COVID-19 confirmed by reverse transcriptase-polymerase reaction chain test (RT-PCR), CXR performed within 24 h (analyzed with two different scores) and recorded prognosis. Clinical and laboratory data were collected. Statistical analysis on the features of 83 in-hospital dead vs 263 recovered patients was performed with univariate (uBLR), multivariate binary logistic regression (mBLR) and ROC curve analysis. uBLR identified significant differences for several variables, most of them intertwined by multiple correlations. mBLR recognized as significant independent predictors for in-hospital mortality age > 75 years, C-reactive protein (CRP) > 60 mg/L, PaO2/FiO2 ratio (P/F) < 250 and CXR "Brixia score" > 7. Among the patients with at least two predictors, the in-hospital mortality rate was 58% against 6% for others [p < 0.0001; RR = 7.6 (4.4-13)]. Patients over 75 years had three other predictors in 35% cases against 10% for others [p < 0.0001, RR = 3.5 (1.9-6.4)]. The greatest risk of death from COVID-19 was age above 75 years, worsened by elevated CRP and CXR score and reduced P/F. Prompt determination of these data at admission to the emergency department could improve COVID-19 pretreatment risk stratification.

Descriptive analysis of a comparison between lung ultrasound and chest radiography in patients suspected of COVID-19.

BACKGROUND: Lung ultrasound (LUS) and chest radiography (CXR) are the most used chest imaging tools in the early diagnosis of COVID-19 associated pneumonia. However, the relationship between LUS and CXR is not clearly defined. The aim of our study was to describe the comparison between LUS interpretation and CXR readings in the first approach to patients suspected of COVID-19. METHODS: In the time of the first COVID-19 pandemic surge, we prospectively evaluated adult patients presenting to an emergency department complaining of symptoms raising suspicion of COVID-19. Patients were studied by LUS and only those performing also CXR were analyzed. All the patients performed viral reverse transcriptase-polymerase chain reaction (RT-PCR). LUS studies were classified in 4 categories of probabilities, based on the presence of typical or alternative signs of COVID-19-associated interstitial pneumonia. Accordingly, the CXR readings were retrospectively adapted by 2 experts in 4 categories following the standard language that describes the computed tomography (CT) findings. Patients were divided in two groups, based on the agreement of the LUS and CXR categories. Results were also compared to RT-PCR and, when available, to CT studies. RESULTS: We analyzed 139 cases (55 women, mean age 59.1 ± 15.5 years old). The LUS vs CXR results disagreed in 60 (43.2%) cases. RT-PCR was positive in 88 (63.3%) cases. In 45 cases, a CT scan was also performed and only 4 disagreed with LUS interpretation versus 24 in the comparison between CT and CXR. In 18 cases, LUS detected signs of COVID-19 pneumonia (high and intermediate probabilities) while CXR reading was negative; in 14 of these cases, a CT scan or a RT-PCR-positive result confirmed the LUS interpretation. In 6 cases, LUS detected signs of alternative diagnoses to COVID-19 pneumonia while CXR was negative; in 4 of these cases, CT scan confirmed atypical findings. CONCLUSION: Our study demonstrated a strong disagreement between LUS interpretation and CXR reading in the early approach to patients suspected of COVID-19. Comparison with CT studies and RT-PCR results seems to confirm the superiority of LUS over a second retrospective reading of CXR.

Baseline chest X-ray in coronavirus disease 19 (COVID-19) patients: association with clinical and laboratory data.

PURPOSE: To assess the reliability of CXR and to describe CXR findings and clinical and laboratory characteristics associated with positive and negative CXR. METHODS: Retrospective two-center study on consecutive patients admitted to the emergency department of two north-western Italian hospitals in March 2020 with clinical suspicion of COVID-19 confirmed by RT-PCR and who underwent CXR within 24 h of the swab execution. 260 patients (61% male, 62.8 ± 15.8 year) were enrolled. CXRs were rated as positive (CXR+) or negative (CXR-), and features reported included presence and distribution of airspace opacities, pleural effusion and reduction in lung volumes. Clinical and laboratory data were collected. Statistical analysis was performed with nonparametric tests, binary logistic regression (BLR) and ROC curve analysis. RESULTS: Sensitivity of CXR was 61.1% (95%CI 55-67%) with a typical presence of bilateral (62.3%) airspace opacification, more often with a lower zone (88.7%) and peripheral (43.4%) distribution. At univariate analysis, several factors were found to differ significantly between CXR+ and CXR-. The BLR confirmed as significant predictors only lactate dehydrogenase (LDH), C-reactive protein (CRP) and interval between the onset of symptoms and the execution of CXR. The ROC curve procedure determined that CRX+ was associated with LDH > 500 UI/L (AUC = 0.878), CRP > 30 mg/L (AUC = 0.830) and interval between the onset of symptoms and the execution of CXR > 4 days (AUC = 0.75). The presence of two out of three of the above-mentioned predictors resulted in CXR+ in 92.5% of cases, whereas their absence in 7.4%. CONCLUSION: CXR has a low sensitivity. LDH, CRP and interval between the onset of symptoms and the execution of CXR are major predictors for a positive CXR.