How automated measurement of lung lesions can predict the course of patients hospitalized with SARS-Cov-2 pneumonia

doi:10.21203/rs.3.rs-1705712/v1

Download PDF

Article

How automated measurement of lung lesions can predict the course of patients hospitalized with SARS-Cov-2 pneumonia

https://doi.org/10.21203/rs.3.rs-1705712/v1

This work is licensed under a CC BY 4.0 License

Version 1

posted

You are reading this latest preprint version

Introduction

The aim of this study was to evaluate if an automated measurement of lung lesions, epicardial fat and pericardial volume during the days surrounding hospital admission for COVID-19 pneumonia may predict intubation or mortality. The second purpose of this study was to assess whether the association of these Computed Tomography (CT) measures with the SOFA (Sequential Organ Failure Assessment) score, could predict intubation and mortality better than the SOFA score alone.

Methods

This observational retrospective study was conducted in Timone university hospital in Marseille in France, between March 10^th and May 10^th 2020. All adult patients with COVID-19, admitted with respiratory symptoms and having performed a chest CT three days before to two days after admission were eligible for inclusion. All chest CTs were analyzed using a local automated CT measurement software. The primary outcome was invasive mechanical ventilation (IMV) or death during the 60-day follow-up. Wilcoxon-Mann-Whitney test was used for univariate analysis and logistic regression were calculated for multivariate analysis.

Results

176 patients were included in the study. 57 (32.4%) received IMV or died during the 60-day follow-up. After univariate analysis, all lung automated volumetric measures of ground-glass (p=0.015), consolidation (p<0.001) and all lesions to parenchymal volume ratio (p<0.001) were significantly higher for the patients who required IMV or who died. All pulmonary-lesion rate was tested in multivariate analysis and remained significantly higher in the IMV or death group (p=0.003), with an Odd Ratio of 3.52 (1.55-8.01, 95% CI) for patients who had more than 19.5% of pulmonary lesion. Pericardial volume and epicardial fat were not significantly associated with IMV or mortality. In this study, the association of the criterion “pulmonary lesion >20%” to the SOFA score improves its predictive value on IMV or mortality with a AUC of 0.82.

Conclusion

Automated chest CT measures of COVID-19 patients with respiratory symptoms admitted to hospital showed a significantly higher rate of lung lesions (ground glass, consolidation, or both) for those who later died or required IMV. Furthermore, the association of these automated CT measures to the SOFA score could help select patients requiring ICU upon entering hospital.

COVID-19

SARS-Cov-2

Prognostic factors

Computed tomography

Automated volumetric measures

The epidemic waves of COVID-19 have punctuated hospital activities around the globe. During each crisis, the management of hospital beds remains a central concern. In this context, the hospitalization of patients in sectors adapted to their severity is fundamental in order not to saturate intensive care units.

At this time, number of articles have identified the main gravity factors and some have tried to create scores that could predict bad evolution of the disease (1,2,3,4). Moreover, performing a chest computed tomography (CT) during the first days of hospitalization in patients with COVID-19 has become a habit for most practitioners, allowing to comfort the diagnosis of COVID-19 (Kovacs A el al. Eur Radiol 2021), assess the lesion extent and diagnose eventual emboly. It appears that ground-glass opacities (GGO) and consolidations are classical lesions of COVID-19 (5,6,7). Some studies tried to use CT quantification of the lung lesions to predict gravity (8,9,10,11), but the actual prognostic value of CT-scan in severe COVID-19 is still a matter of debate. Most of these studies were Chinese, and few of them used an automated lesion quantification software (12,13). Some studies also suggested that visceral adiposity and particularly pericardial fat could be a gravity factor of COVID-19 by serving as a reservoir for the virus and increasing the inflammatory response (13,14).

Thus, the aim of our study was to evaluate if an automated measurement of lung lesions, epicardial fat and pericardial volume at an early stage of COVID-19 could be predictive of mortality and invasive mechanical ventilation (IMV).

Patients characteristics

Between March 10th, 2020 and May 10th, 2020, 259 patients were hospitalized in Timone University Hospital for COVID-19 suspicion. Among those patients, 176 had confirmed COVID-19 infection and respiratory symptoms with a chest CT performed 3 days before to 2 days after hospital admission. 57 of them required IMV or died during the 60-day follow-up, all during the first 19 days of hospitalization (figure 1). Their characteristics are reported in Table 1. At hospital admission, the median age of hospitalized patients was 64 [IQ 54-72] years old with 67 women (38.1%) and 109 men (61.9%). Their most frequent comorbidities were hypertension 84/176 (47.7%), overweight 80/176 (45.5%), diabetes 42/176 (23.9%) and dyslipidemia 38/176 (21.6%). On univariate analysis, age was significantly associated with IMV or death (p<0.001). The significant comorbidities associated with IMV or mortality were hypertension (OR 2.53, 95%CI 1.32-4.86, p<0.005), diabetes (OR 2.72, 95%CI 1.33-5.56, p<0.005), metabolic syndrome (OR 2.9, 95%CI 1.2-6.96, p=0.014) and chronic kidney disease (OR 4.69, 95%CI 1.35-16.31, p=0.02).

Lymphopenia was commonly observed, with a median value of 0.9 x10⁹/L in the survivors group against 0.8 x10⁹/L in the IMV or death group (p=0.028). The NLR was also significantly different in the two groups with a median value at 4.8 in the survivors group against 8.7 in the IMV or death group (p<0.001). Hemoglobin and albumin were significantly lower in the IMV or death group (p=0.011 and p=0.001 respectively) whereas CRP, creatinine and troponin were significantly higher (p=0.002, p=0.013 and p=0.01 respectively) than in the survivors group.

Concerning the symptoms before admission, anosmia and ageusia were the only significant factors associated with good prognosis (OR 0.15, 95%CI 0.03-0.68, p=0.006).

Automated CT measurements

Automated volumetry of the chest CT measured a median total lesion volume at 14.6% [IQ 5.9-27.1] of the parenchyma, with a GGO median at 11% [IQ 4-19] and a consolidation median at 2.2 % [IQ 0.4-5.1] of lung parenchyma. All these measures were significantly different between the survivors group and the IMV or death group on the univariate analysis with a median total lesion volume of 11.7% [IQ 4.6-20.9] for the survivors group and 21.6% [IQ 10.3-34.8] for the IMV or death group (p <0.001); a median GGO of 9% [IQ 3-17] of lung volume in the survivors group versus 15% [IQ 8-27] in the IMV or death group (p<0.001) and a median pulmonary condensation of 1.4% [IQ 0.3-3.8] in survivors group against 4.2% [IQ 1-8.7] in the IMV or death group (p <0.001). On the other hand, no significant difference was found between the two groups concerning the measures of pericardial volume and epicardial fat volume.

The univariate analysis on IMV shows a average of 15.9% of pulmonary lesions for non-intubated patients against 25.6% for intubated patients (p <0.001).

The univariate analysis on mortality alone, shows no significant difference for pulmonary lesions between the surviving patients (22,3%) and the deceased ones (17,1%) (p = 0.101).

The ROC curve of the total lung lesions rate showed an optimal discriminant cut-off at 19.5% for IMV and mortality. In the multivariate analysis without NLR, age and total lung lesions rate remained significantly different in the two groups with an OR at 1.45 (1.07-1.96, 95% CI, p=0.018) for each ten years of age, and an OR at 3.52 (1.55-8.01, 95% CI, p=0.003) for all patients who had more than 19.5% of lung lesions. In the analysis with NLR, NLR remained the only significant variable with an OR at 1.126 (1.021-1.241 95% CI, p=0.017). Those results are reported in Table 2.

Severity of COVID-19 and complications

70 of the 176 patients (39.8%) went to ICU during the 60 days follow-up. 34 (19.3%) of them needed vasopressor, 12 (6.9%) suffered from thrombosis or embolism, 6 (3.4%) needed extracorporeal mechanical oxygenation and 6 (3.4%) needed renal replacement therapy.

15 days post admission, 98 (55.7%) patients were alive and got discharged from the hospital, 30 (17%) still needed high flow oxygen, non invasive ventilation or IMV in ICU and 21 (11.9%) had died (Table 3). In addition, we observe that all patients who required at least low flow oxygen or more had on average of 23.6 % pulmonary damage against 14.7 % in patients who did not (p<0.001).

Prediction of IMV or death in COVID-19

The Area under the curve (AUC) of ROC curves showed that the best predictions for IMV or death were obtained with the SAPS II score and the SOFA score, with an AUC of 0.81 and 0.80 respectively. For the clinical values on admission, the best AUC were obtained with the SpO2 to FiO2 ratio and the temperature on admission with an AUC at 0.79 and 0.78 respectively. For CT volumetry, the lesions/pulmonary volume ratio had an AUC of 0.68 (Figure 2). We created a composite score by adding the criterion “more than 20% CT pulmonary lesion” to the SOFA score. We chose 20% because the ROC curve of the pulmonary lesion to lung volume ratio showed a best predictive cut-off at 19.5%. This upgraded SOFA score’s AUC to 0.82, with an ideal predictive cut-off >2 (sensitivity 73.7%, specificity 79.8%), making it a better prognostic score than SAPS II while being very simple in clinical practice.

The results of the present study show that the extent of pulmonary lesions (Consolidation and GGO) assessed by chest CT obtained during the days surrounding hospital admission of patients with COVID-19 is associated to the prognosis, independently of other demographic and biological parameters.

Those results remain significant when comparing consolidation or GGO independently, which suggests that the residual well ventilated lung volume conditions the prognosis. On the other hand, measures of pericardial volume and epicardial fat did not show a significant prognostic impact in our study, unlike in other studies (13,14). However, those studies focused on younger patients and showed that pericardial volume and epicardial fat where associated to an increased myocarditis risk.

In this study, univariate analyses focusing specifically on IMV or on mortality show that the extent of lung lesions is significantly correlated with the risk of IMV but not with the risk of mortality. This can be explained by the fact that the patients with lung damages survived thanks to IMV. The authors of this study believe that if the study was conducted in a center without an ICU or if there were more participants in this study, the rate of lung injury would probably have remained significantly associated with mortality.

The use of automated measurement of lung volume and lung lesions is the particularity of this study. Studies comparing the accuracy of computerized measurements of lung lesions in the context of COVID-19 have shown a similar efficiency compared to measurements realized by experienced radiologists (21,22).

Moreover, the ROC curve of our study suggests a cut-off of approximately 20% lung lesions over total parenchymal volume for the prediction of IMV and mortality. It can be noted that patients who still need IMV or at least low flow oxygen after 15 days of hospitalization have on average 23.6% lung lesions on the initial chest CT. However, it should also be noted that the AUC of pulmonary lesion rate is only 0.68 with a sensitivity of 58% and a specificity of 73%. This CT data must therefore be interpreted along with other clinical and biological parameters for a better prognosis estimation.

In this regard, our study confirms in univariate analysis various gravity factors which had already been demonstrated in previous studies, such as age, comorbidities (hypertension, metabolic syndrome, diabetes), but also some biological parameters such as NLR, CRP and Troponin elevation on admission. In multivariate analysis, only age remained significantly associated with a higher risk of IMV and mortality, probably due to the study’s small sample size. Despite the lack of data regarding the NLR and because it appeared to be relevant in several previous studies (18,19,20), we performed an additional multivariate analysis adding the NLR. It remained significant despite the low residual power, suggesting a strong pronostic impact of lymphopenia. Concerning anosmia-ageusia, although it was significantly associated with good prognosis in this study, we believe it was subject to reporting bias, as physicians did not investigate whether this symptom was present in the most severe patients.

One of the originalities of this study is also that it evaluated numerous prognostic scores on hospital arrival. It appears that the simplest and most effective score for assessing the risk of intubation and mortality from COVID-19 in the emergency room is the SOFA score, widely used in ICUs to predict outcomes in patients with sepsis (Vincent JL et al. Intensive Care Med 1996). In our study, the SOFA ROC curve showed an AUC of 0.80 with an optimal prediction cut-off > 2 with a sensitivity of 60% and a specificity of 92%. Although the SOFA score was reported to be imprecise for predicting mortality in patients with COVID-19 already necessiting IMV (Raschke RA et al. JAMA 2021), it seems to be an interesting prognostic score when used early in hospitalized COVID-19 patients who are not under IMV. We can see in our study that by adding the criterion “>20% pulmonary lesion” to the SOFA score, we improve its predicting value with AUC of 0.82 associated to a best prediction cut-off > 2 (sensitivity of 73.7%, specificity of 79.8%). This modified SOFA score should ideally be tested with another cohort.

The ROX index, which is a simple and exclusively clinical score firstly described by Roca et al in 2016 (23), also has an interesting prognostic performance with an AUC of 0.78 and an optimal prediction cut-off at 12 (sensitivity of 62%, specificity of 87%).

Our study has limitations. We arbitrarily chose to select all the patients who had chest CT performed during the five days surrounding hospitalization in order to include as many patients as possible. This does not take into account the date of symptom onset and we have no evidence that pulmonary lesions do not significantly change in five days. This was a monocentric study and differences in medical treatment were not systematically analyzed. A large proportion of patients received hydroxychloroquine with no significant difference between the survivors group and those who died or had IMV.

In conclusion, in our retrospective cohort of 176 COVID-19 patients hospitalized with respiratory symptoms, the automated volumetric measures of chest CT performed during the days surrounding the hospital admission showed a significantly higher rate of pulmonary GGO and consolidation in patients who required IMV or who died during the 60-day follow up. Therefore, this data should be interpreted with other clinical and biological data in order identify high-risk patients. However, pericardial volume and epicardial fat volume were not significantly associated to a higher risk of IMV or mortality. The combination of the SOFA score with pulmonary lesion > 20% can simply and efficiently provide help for the triage of these patients.

Patient population

In this observational, monocentric, retrospective study, data were collected from all adult patients with COVID-19 admitted for respiratory symptoms and who received chest computed tomography three days before to two days after admission at Timone Hospital in Marseille, from March 10th, 2020 to May 10th, 2020. Patients were considered COVID-19 positive if they had laboratory confirmation for SARS-CoV-2 defined as positive result of real-time reverse transcriptase-polymerase chain reaction (RT-PCR) assay of nasal or pharyngeal swab, or if they had respiratory symptoms and a compatible chest CT after being exposed to confirmed infected individuals. Indeed, some studies have found that chest CTs has better sensitivity to identify COVID-19 pneumonia than RT-PCR (15,16). The exclusion criteria were: patients without respiratory symptoms or patients already under mechanical ventilation on admission. Survival status up to 2 months after hospital admission was available for all of them.

The ethical committee of the French Society of Anesthesia and Intensive Care Medicine (SFAR) and the local committee of Marseille university (AMU) approved this study. The need of informed consent was waived by Marseille University Ethics Committee (2020-53) and the French Society of Anaesthesia and Intensive Care Medicine (00010254-2020-06), and all aspect of the work were carried out in accordance with relevant guidelines/regulations including the Helsinki Declaration.

Outcomes

The primary outcome was the decision of mechanical ventilation with oro-tracheal intubation or death during the 60-day follow-up. Patients were divided into two groups depending on the outcome: patients who survived two months after admission without the need of IMV (hospitalized in conventional units or in ICU) were considered as the survivors group, and the other patients were considered as the IMV or death group.

Secondary outcomes included IMV and mortality separately, clinical status at day 15 using a six points ordinal scale adapted from the World Health Organization R&D Blueprint expert group (1 = alive and discharged of hospital, 2 = hospitalized without oxygen use, 3 = hospitalized using low flow oxygen, 4 = hospitalized using non invasive ventilation or high flow nasal cannula, 5 = hospitalized, using IMV and 6 = dead).

Chest CT protocol and deep learning pipeline

All patients underwent Low dose computed tomography in deep inspiration. All imaging data were reconstructed using high resolution and standard algorithms.

To quantify the lesions, we used an internal deep learning pipeline developed by Dr Bartoli and al and submitted to publication. It uses convolutional neural network (CNN) and allows a fully automated segmentation of COVID-19 pulmonary lesions, epicardial fat and pericardial volume. It is composed of three 2D slice-based CNN models. The first (CN1) is a classification network that enables to distinguish the slices that contains part of the lungs, the pericardial volume and the epicardial fat from the rest. The second and third models (CN2 and CN3) were U-Net inspired segmentation networks. CN2 is trained to segment slices into GGO, consolidation or normal lung, epicardial fat and pericardial volume. CN3 has the same role as CN2 and served has a refiner.

Data collection

Data was obtained retrospectively for each patient from their electronic charts using local software Axigate 5.8.0P10. The collected baseline characteristics were age, sex, overweight (BMI > 25 kg/m²), comorbidities and long-term medications. We identified patients with metabolic syndrome using the definition of the International Diabetes Federation (17). Severity of illness scores such as SAPS II (Simplified Acute Physiology Score II), SOFA (Sequential Organ Failure Assessment), APACHE II (Acute Physiology and Chronic Health Evaluation II), Charlson comorbidity index, MuLBSTA score, CURB-65 score and ROX index were calculated based on the data from the first 24 hours after hospital admission. When patients underwent flow oxygen therapy, FiO2 was estimated as follows : FiO2 = 21% + 3 x L (L = number of oxygen liter). Significant events and complications such as thrombosis and embolism, secondary infections, severe bleeding, need for vasopressors, renal replacement therapy or extracorporeal mechanical oxygenation were reported. For laboratory results, we took the first available values of the first 48 hours post hospital admission. Limitation on invasive ventilation for ethical reasons, was also reported. Finally, we took into consideration the onset of symptoms, the day of admission at the hospital, the day of death or hospital discharge and the vital status at the end of the 60-day follow up. For this last information, we contacted hospitals or rehabilitation centers by phone call if the patients had already left our institution.

Patient management in services and ICU, supportive measures, use of pharmaceuticals and decision of oro-tracheal intubation were performed at the discretion of the treating physicians.

Statistical analysis

Statistical analyses were performed using IBMSPSS STatistics version 20. Categorical variables were reported as frequencies (with corresponding percentages) and continuous variables medians (with interquartile ranges). The median values were compared with Wilcoxon-Mann-Whitney test and categorical variables were compared using Pearson Chi-square test or Fisher’s exact test as appropriate. Logistic regression models were calculated for multivariate analysis. Statistical significance was defined as p-value < 0,05.

We conducted a univariate analysis on the primary outcome (IMV or death), as well as one on death and one on IMV. Then we conducted multivariate analysis on primary outcome using 7 significant and unrelated variables : Age, Hypertension, Metabolic syndrome, CT pulmonary lesions to pulmonary volume ratio, Hemoglobin, CRP, Creatinine. Even though Neutrophil to Lymphocyte ratio (NLR) had multiple missing data, we also decided to conduct another multivariate analysis adding the NLR to the 7 variables because it seemed to be relevant on gravity evolution on some preliminary studies (18,19,20).

For the continuous variables, we also modelized ROC curves in order to evaluate the best predictive scores, CT measures or biological variables for predicting IMV or death. We used the ROC curve of the total lung lesions rate to identify an optimal discriminant cut-off and we analyzed this cut-off in multivariate analysis as well.

Acknowledgments: We thank J. Albanese, N. Bruder, P. Chanez, M. Gainnier, J.R. Harle, L. Papazian, J.M. Forel, C. Guidon, V. Veit, L. Zieleskiewicz for collecting cases. We gratefully acknowledge all the health care workers on the front line and all the patients involved in the study.

Author Contributions: Y.B. and D.T. contributed to the conception and design of the study, had full access to all the data in the study and take responsibility for the integrity of the data and the accuracy of the data analysis. D.T., M.V., Y.B., J.B., H.M., J.C., M.B., N.P., A.B., N.P., G.A., D.L., A.J., V.B. and B.P. contributed to the acquisition of data. J.B., D.T., G.A., M.L., L.V., N.B. and Y.B. contributed to the analysis and interpretation of the data. A.L. and L.B. contributed to the statistical analyses. All authors participated in manuscript writing, revision and approval for final submission. Y.B. and D.T. contributed equally and shared first authorship. All authors have read and agreed to the published version of the manuscript.

Funding: The authors receive no funding for this work.
Institutional Review Board Statement: This study was approved by the local Ethics Commission

(2020-53) and the French Society of Anaesthesia and Intensive Care Medicine (00010254-2020-06).

Informed Consent Statement: According to French law, a written informed consent from patients was not necessary, owing to the retrospective nature of the study, but the patients had the possibility to withdraw their health data.

Conflicts of Interest: The authors declare that they have no competing interest.

Wu & McGoogan, « Characteristics of and Important Lessons From the Coronavirus Disease 2019 (COVID-19) Outbreak in China ».
Richardson, Safiya, Jamie S. Hirsch, Mangala Narasimhan, James M. Crawford, Thomas McGinn, Karina W. Davidson, Douglas P. Barnaby, and al. « Presenting Characteristics, Comorbidities, and Outcomes Among 5700 Patients Hospitalized With COVID-19 in the New York City Area ». JAMA 323, n^o 20 (26 may 2020): 205259. https://doi.org/10.1001/jama.2020.6775.
Martin, Salomon, Sandra Fuentes, Catalina Sanchez, Marta Jimenez, Carmen Navarro, Helena Perez, Elena Salamanca, and al. « Development and validation of a laboratory-based risk score to predict the occurrence of critical illness in hospitalized patients with COVID-19 ». Scandinavian Journal of Clinical and Laboratory Investigation 0, n^o 0 (11 may 2021): 18. https://doi.org/10.1080/00365513.2020.1847313.
Petrilli, Christopher M, Simon A Jones, Jie Yang, Harish Rajagopalan, Luke O’Donnell, Yelena Chernyak, Katie A Tobin, Robert J Cerfolio, Fritz Francois, and Leora I Horwitz. « Factors associated with hospital admission and critical illness among 5279 people with coronavirus disease 2019 in New York City: prospective cohort study ». The BMJ 369 (22 may 2020). https://doi.org/10.1136/bmj.m1966.
Ye, Zheng, Yun Zhang, Yi Wang, Zixiang Huang, and Bin Song. « Chest CT manifestations of new coronavirus disease 2019 (COVID-19): a pictorial review ». European Radiology, 19 mars 2020, 19. https://doi.org/10.1007/s00330-020-06801-0.
Caruso, Damiano, Marta Zerunian, Michela Polici, Francesco Pucciarelli, Tiziano Polidori, Carlotta Rucci, Gisella Guido, Benedetta Bracci, Chiara de Dominicis, and Andrea Laghi. « Chest CT Features of COVID-19 in Rome, Italy ». Radiology, 3 avril 2020. https://doi.org/10.1148/radiol.2020201237.
Chung, Michael, Adam Bernheim, Xueyan Mei, Ning Zhang, Mingqian Huang, Xianjun Zeng, Jiufa Cui, and al. « CT Imaging Features of 2019 Novel Coronavirus (2019-nCoV) ». Radiology, 4 february 2020. https://doi.org/10.1148/radiol.2020200230.
Li, Kunhua, Jiong Wu, Faqi Wu, Dajing Guo, Linli Chen, Zheng Fang, and Chuanming Li. « The Clinical and Chest CT Features Associated With Severe and Critical COVID-19 Pneumonia ». Investigative Radiology, 2020. https://doi.org/10.1097/RLI.0000000000000672.
Yuan, Mingli, Wen Yin, Zhaowu Tao, Weijun Tan, and Yi Hu. « Association of radiologic findings with mortality of patients infected with 2019 novel coronavirus in Wuhan, China ». PLoS ONE 15, n^o 3 (19 mars 2020). https://doi.org/10.1371/journal.pone.0230548.
Lyu, Peijie, Xing Liu, Rui Zhang, Lei Shi, and Jianbo Gao. « The performance of chest CT in evaluating the clinical severity of COVID-19 pneumonia: identifying critical cases based on CT characteristics ». Investigative Radiology, 17 april 2020. https://doi.org/10.1097/RLI.0000000000000689.
Meiler, Stefanie, Jan Schaible, Florian Poschenrieder, Gregor Scharf, Florian Zeman, Janine Rennert, Benedikt Pregler, and al. « Can CT performed in the early disease phase predict outcome of patients with COVID 19 pneumonia? Analysis of a cohort of 64 patients from Germany ». European Journal of Radiology 131 (october 2020): 109256. https://doi.org/10.1016/j.ejrad.2020.109256.
Feng, Zhichao, Hui Shen, Kai Gao, Jianpo Su, Shanhu Yao, Qin Liu, Zhimin Yan, and al. « Machine learning based on clinical characteristics and chest CT quantitative measurements for prediction of adverse clinical outcomes in hospitalized patients with COVID-19 ». European Radiology, 15 april 2021, 111. https://doi.org/10.1007/s00330-021-07957-z.
Deng, Ming, Yongjian Qi, Liping Deng, Huawei Wang, Yancheng Xu, Zhen Li, Zhe Meng, Jun Tang, and Zhe Dai. « Obesity as a Potential Predictor of Disease Severity in Young COVID-19 Patients: A Retrospective Study ». Obesity (Silver Spring, Md.), 29 june 2020. https://doi.org/10.1002/oby.22943.
Battisti, Sofia, Claudio Pedone, Nicola Napoli, Emanuele Russo, Vanni Agnoletti, Stefano Geniere Nigra, Caterina Dengo, and al. « Computed Tomography Highlights Increased Visceral Adiposity Associated With Critical Illness in COVID-19 ». Diabetes Care 43, n^o 10 (1 october 2020): e12930. https://doi.org/10.2337/dc20-1333.
Ai, Tao, Zhenlu Yang, Hongyan Hou, Chenao Zhan, Chong Chen, Wenzhi Lv, Qian Tao, Ziyong Sun, and Liming Xia. « Correlation of Chest CT and RT-PCR Testing for Coronavirus Disease 2019 (COVID-19) in China: A Report of 1014 Cases ». Radiology 296, n^o 2 (august 2020): E3240. https://doi.org/10.1148/radiol.2020200642.
Thomas, Clément, Mathieu Naudin, Jean-Pierre Tasu, Charles Leclerc, Lucas Depaire, Marie Subervillle, Mathilde Vionnet, Rémy Guillevin, and Guillaume Herpe. « Efficacy of chest CT scan for COVID-19 diagnosis in a low prevalence and incidence region ». European Radiology, 19 april 2021, 16. https://doi.org/10.1007/s00330-021-07863-4.
Alberti, K. George MM, Paul Zimmet, and Jonathan Shaw. « The Metabolic Syndrome—a New Worldwide Definition ». The Lancet 366, n^o 9491 (24 september 2005): 105962. https://doi.org/10.1016/S0140-6736(05)67402-8.
Alkhatip, Ahmed Abdelaal Ahmed Mahmoud M., Mohamed Gomaa Kamel, Mohamed Khaled Hamza, Ehab Mohamed Farag, Hany Mahmoud Yassin, Mohamed Elayashy, Amr Ahmed Naguib, and al. « The Diagnostic and Prognostic Role of Neutrophil-to-Lymphocyte Ratio in COVID-19: A Systematic Review and Meta-Analysis ». Expert Review of Molecular Diagnostics 0, n^o ja (10 april 2021): null. https://doi.org/10.1080/14737159.2021.1915773.
Simadibrata, Daniel Martin, Julius Calvin, Alya Darin Wijaya, and Naufal Arkan Abiyyu Ibrahim. « Neutrophil-to-lymphocyte ratio on admission to predict the severity and mortality of COVID-19 patients: A meta-analysis ». The American Journal of Emergency Medicine 42 (april 2021): 6069. https://doi.org/10.1016/j.ajem.2021.01.006.
Li, Xiaoming, Chao Liu, Zhi Mao, Minglu Xiao, Li Wang, Shuang Qi, and Feihu Zhou. « Predictive values of neutrophil-to-lymphocyte ratio on disease severity and mortality in COVID-19 patients: a systematic review and meta-analysis ». Critical Care 24 (16 november 2020). https://doi.org/10.1186/s13054-020-03374-8.
Colombi, Davide, Flavio C. Bodini, Marcello Petrini, Gabriele Maffi, Nicola Morelli, Gianluca Milanese, Mario Silva, Nicola Sverzellati, and Emanuele Michieletti. « Well-aerated Lung on Admitting Chest CT to Predict Adverse Outcome in COVID-19 Pneumonia ». Radiology, 17 april 2020. https://doi.org/10.1148/radiol.2020201433.
Li, Yuehua, Kai Shang, Wei Bian, Li He, Ying Fan, Tao Ren, and Jiayin Zhang. « Prediction of disease progression in patients with COVID-19 by artificial intelligence assisted lesion quantification ». Scientific Reports 10 (16 december 2020). https://doi.org/10.1038/s41598-020-79097-1.
Roca, Oriol, Jonathan Messika, Berta Caralt, Marina García-de-Acilu, Benjamin Sztrymf, Jean-Damien Ricard, and Joan R. Masclans. « Predicting Success of High-Flow Nasal Cannula in Pneumonia Patients with Hypoxemic Respiratory Failure: The Utility of the ROX Index ». Journal of Critical Care 35 (1 october 2016): 200205. https://doi.org/10.1016/j.jcrc.2016.05.022.
Ronneberger, Olaf, Philipp Fischer, and Thomas Brox. « U-Net: Convolutional Networks for Biomedical Image Segmentation ». arXiv:1505.04597 [cs], 18 may 2015. http://arxiv.org/abs/1505.04597.
Khened, Mahendra, Varghese Alex Kollerathu, and Ganapathy Krishnamurthi. « Fully Convolutional Multi-Scale Residual DenseNets for Cardiac Segmentation and Automated Cardiac Diagnosis Using Ensemble of Classifiers ». Medical Image Analysis 51 (1 january 2019): 2145. https://doi.org/10.1016/j.media.2018.10.004.
Ruder, Sebastian. « An overview of gradient descent optimization algorithms ». arXiv:1609.04747 [cs], 15 june 2017. http://arxiv.org/abs/1609.04747.

Table 1-3 are available in the Supplemental Files section.

No competing interests reported.

Tables.docx

Download PDF

Version 1

posted

You are reading this latest preprint version

How automated measurement of lung lesions can predict the course of patients hospitalized with SARS-Cov-2 pneumonia

Status:

Version 1

Abstract

Figures

Introduction

Results

Discussion

Methods

Patient population

Outcomes

Chest CT protocol and deep learning pipeline

Data collection

Statistical analysis

Declarations

References

Tables

Additional Declarations

Supplementary Files

Status:

Version 1