|Year : 2022 | Volume
| Issue : 4 | Page : 136-144
Dynamic chest computed tomography change analysis and prediction of length of stay for delta variant COVID-19 patients
Xiaoyan Xin1, Wen Yang1, Ying Wei2, Jun Hu1, Xin Peng1, Yi Sun3, Cong Long1, Xin Zhang1, Chao Du4, Feng Shi2, Bing Zhang1
1 Department of Radiology, Nanjing Drum Tower Hospital, The Affiliated Hospital of Nanjing University Medical School, Jiangsu, China
2 Department of Research and Development, Shanghai United Imaging Intelligence Co., Ltd., Xuhui, Shanghai, China
3 Department of Radiology, Nanjing Drum Tower Hospital Clinical College of Nanjing University of Chinese Medicine, Jiangsu, China
4 Department of Radiology, The Second Affiliated Hospital of Nanjing University of Chinese Medicine, Nanjing, China
|Date of Submission||29-Sep-2022|
|Date of Decision||28-Oct-2022|
|Date of Acceptance||03-Dec-2022|
|Date of Web Publication||21-Mar-2023|
Department of Radiology, Nanjing Drum Tower Hospital, The Affiliated Hospital of Nanjing University Medical School, Nanjing 210008, Jiangsu
Department of Research and Development, Shanghai United Imaging Intelligence Co., Ltd., 701 Yunjin Road, Xuhui District, Shanghai, 200232
Source of Support: None, Conflict of Interest: None
OBJECTIVE: As hospital admission rate is high during the COVID-19 pandemic, hospital length of stay (LOS) is a key indicator of medical resource allocation. This study aimed to elucidate specific dynamic longitudinal computed tomography (CT) imaging changes for patients with COVID-19 over in-hospital and predict individual LOS of COVID-19 patients with Delta variant of SARS-CoV-2 using the machine learning method.
MATERIALS AND METHODS: This retrospective study recruited 448 COVID-19 patients with a total of 1761 CT scans from July 14, 2021 to August 20, 2021 with an averaged hospital LOS of 22.5 ± 7.0 days. Imaging features were extracted from each CT scan, including CT morphological characteristics and artificial intelligence (AI) extracted features. Clinical features were obtained from each patient's initial admission. The infection distribution in lung fields and progression pattern tendency was analyzed. Then, to construct a model to predict patient LOS, each CT scan was considered as an independent sample to predict the LOS from the current CT scan time point to hospital discharge combining with the patients' corresponding clinical features. The 1761 follow-up CT data were randomly split into training set and testing set with a ratio of 7:3 at patient-level. A total of 85 most related clinical and imaging features selected by Least Absolute Shrinkage and Selection Operator were used to construct LOS prediction model.
RESULTS: Infection-related features were obtained, such as the percentage of the infected region of lung, ground-glass opacity (GGO), consolidation and crazy-paving pattern, and air bronchograms. Their longitudinal changes show that the progression changes significantly in the earlier stages (0–3 days to 4–6 days), and then, changes tend to be statistically subtle, except for the intensity range between (−470 and −70) HU which exhibits a significant increase followed by a continuous significant decrease. Furthermore, the bilateral lower lobes, especially the right lower lobe, present more severe. Compared with other models, combining the clinical, imaging reading, and AI features to build the LOS prediction model achieved the highest R2 of 0.854 and 0.463, Pearson correlation coefficient of 0.939 and 0.696, and lowest mean absolute error of 2.405 and 4.426, and mean squared error of 9.176 and 34.728 on the training and testing set.
CONCLUSION: The most obvious progression changes were significantly in the earlier stages (0–3 days to 4–6 days) and the bilateral lower lobes, especially the right lower lobe. GGO, consolidation, and crazy-paving pattern and air bronchograms are the most main CT findings according to the longitudinal changes of infection-related features with LOS (day). The LOS prediction model of combining clinical, imaging reading, and AI features achieved optimum performance.
Keywords: Chest computed tomography, COVID-19, delta variant, length of stay, SARS-CoV-2
|How to cite this article:|
Xin X, Yang W, Wei Y, Hu J, Peng X, Sun Y, Long C, Zhang X, Du C, Shi F, Zhang B. Dynamic chest computed tomography change analysis and prediction of length of stay for delta variant COVID-19 patients. Radiol Infect Dis 2022;9:136-44
|How to cite this URL:|
Xin X, Yang W, Wei Y, Hu J, Peng X, Sun Y, Long C, Zhang X, Du C, Shi F, Zhang B. Dynamic chest computed tomography change analysis and prediction of length of stay for delta variant COVID-19 patients. Radiol Infect Dis [serial online] 2022 [cited 2023 Jun 3];9:136-44. Available from: http://www.ridiseases.org/text.asp?2022/9/4/136/372195
| Introduction|| |
The COVID-19 pandemic is the worst infectious disease to occur globally in a century. The World Health Organization (WHO) lists the Alpha variant (first identified in the UK), Beta (first in South Africa), Gamma (in Brazil), and Delta (in India) as variants of “concern.” Patients infected with the delta variant of SARS-CoV-2, the novel coronavirus B.1.617.2 variant was announced by the WHO on May 11, 2021, have also been found in many provinces in China. To effectively control the chain of transmission, isolation and support for treatment are critical. The space for isolation and medical care to patients with COVID-19 give a great burden and pressure on hospitals and medical staff. Therefore, the ward occupancy has become a critical concern. Length of stay (LOS) of COVID-19 patients measures the length of time elapsed between a patient's hospital admittance and discharge, which is a valuable indicator in optimizing the effectiveness of medical resource allocation. Hence, predicting hospital LOS to estimate how long each patient will require inpatient care is useful for clinical decision-making and contingency planning.
The hospital LOS is associated with the disease severity, the level of care required, and the geographic setting due to varying COVID-19 care guidelines. For instance, some patient characteristics (e.g., age, fever, vaccination status, and comorbidities) may affect the disease severity, which are likely to influence duration and LOS., Therefore, clinical features, imaging findings, and nucleic acid tests are all important for the LOS with COVID-19 patient.
Literature have studied that a number of biomarkers are considered potential predictors of the severity and LOS for patients with COVID-19. LOS was longer in patients receiving Remdesivir. Elevated monocyte distribution width (MDW) was associated with a prolonged LOS. Lin et al. have designed an AI-based model to predict the LOS for COVID-19 patients based. The model used clinical features (such as fever, tachypnea, and MDW) of admitted COVID-19 patients to estimate whether the LOS is beyond 14 days or not, which is relative rough for accurate LOS prediction. Chest computed tomography (CT) has been proven to play an important role in the diagnosis and disease course detection of COVID-19. Advanced artificial intelligence (AI) techniques have been successfully performed in COVID-19 detection and classification tasks for patient benefit and epidemic prevention. However, studies on CT findings and LOS have not been reported.
In this study, taking advantage of large-scale longitudinal data, we aim to investigate specific COVID-19-related imaging features and explore their dynamic changes over the LOS, and combined the baseline clinical features to build a predictive model that could achieve simultaneous prediction of LOS from the current CT scan time point to hospital discharge.
| Materials and Methods|| |
This retrospective study was approved by the ethics committees of the Second Affiliated Hospital of Nanjing University of Chinese Medicine, Nanjing, China. The requirement to obtain informed consent was waived by the review board. All data has been anonymized prior to analysis.
A total of 504 patients infected with the Novel Coronavirus B.1.617.2 variant of COVID-19 were retrospectively included in this study from July 14, 2021 to August 20 in the Second Affiliated Hospital of Nanjing University of Chinese Medicine, Nanjing, China.
All patients were confirmed to have COVID-19 following the Diagnosis and Treatment Protocol of Pneumonia in Novel Coronavirus Infection (Trial Version 8) of the National Health Commission of the People's Republic of China. The inclusion criteria were as follows: (a) had been confirmation of a pneumonia etiology based on positive result of polymerase chain reaction (PCR) from swab tests for COVID-19 patients; (b) belonged to B.1.617.2 mutant by high-throughput whole-genome sequencing; (c) have decent quality CT scans; and (d) electronic records were available. Conversely, patients were excluded from the study if they met any of the following exclusion criteria: (a) had symptoms of pneumonia not caused by COVID-19, such as chronic obstructive pulmonary disease, (b) had an unconfirmed etiology of pneumonia; (c) had no CT scans available; and/or (d) had an unconfirmed vaccination status.
According to the COVID-19 Diagnosis and Treatment Protocol (the Eighth Edition) of China, the criteria for a COVID-19 patient to be discharged from hospital is as follows: (1) 3 days after his/her body temperature has returned to normal; (2) respiratory symptoms improved significantly; (3) lung imaging showed significant improvement in acute exudative lesions; and (4) two consecutive negative nucleic acid tests of respiratory tract specimens (at least 24 h between samples).
Finally, total of 448 patients with clinical data and 1761 follow-up CT scans were retrospectively included. CT images were acquired during the patients' admission. Here the illness day 0 represents the initial admitted time that the patient was diagnosed as COVID-19 with positive PCR or highly suspected of COVID-19.
All patients were undergoing chest CT scanning with a Multi-Detector Computed Tomography (16-MDCT) scanner (Aquilion TSX101A, Toshiba, Tokyo, Japan) or a 64-MDCT scanner (Brilliance 64, Philips Healthcare, Cleveland, Ohio, USA). All chest CT scans were undergoing with the patient in the supine position. Breath-hold training was carried out before each examination. All patients were asked to hold their breath at the end of inspiration as long as possible. The following parameters were used: a tube voltage of 120 kVp, automatic tube current modulation technique, a reconstruction matrix of 512 × 512 pixels, a section thickness of 0.75 mm, and an interval of 1.5 mm.
Clinical and computed tomography morphological characteristic evaluation
Clinical characteristics, such as gender, age, fever, cough and laboratory indicators, were collected from the electronic medical records at the hospital. The CT morphological characteristics (i.e., image reading feature) were summarized as categorical variables.
For the CT morphological characteristics, all CT images were reviewed retrospectively by two primary radiologists (W.Y and Q.Q.F), with approximately 6 and 2 years of experience in thoracic imaging respectively, especially in viral pneumonias. The radiologists reviewed the images independently. When the primary radiologists hold different opinions, a third chest radiologist with 10 years of experience (X.Y.X) would adjudicate the final decision.
CT images with following characteristics were defined as parenchymal abnormalities: (a) presence of ground-glass opacities, (b) presence of consolidation, (c) presence of fibrous strands, (d) presence of crazy-paving pattern, (e) presence of air bronchogram, (f) presence of cavity, (g) presence of a pleural effusion, (h) presence of mediastinal enlarged lymph nodes (defined as lymph node size of 10 mm in short-axis dimension), and (i) presence of pulmonary fibrosis. This CT image revision did not include abnormal lobe distribution and extent of involvement.
Artificial intelligence feature extraction
A total of 154 quantitative handcrafted lung burden features (i.e., AI features) were extracted from each CT image of COVID-19 patient., Briefly, the infected lung regions and bilateral lung fields were automatically segmented with the uAI research portal, which was previously evaluated showing a high agreement with manual annotation with dice of 91.6%. In this study, all images were automatically segmented to extract the lung, lung lobes, pulmonary segments, and infected lung regions. After that, 154 handcrafted features were extracted from each CT image in total. The handcrafted features are divided into six groups based on different characteristics of the lesions, including density, volume, mass, infected lesion number, surface area, and histogram distribution. The details of handcrafted features have been described in prior publication. Meanwhile, we also calculated CT score, a widely used evaluation method by radiologists, to quantify the extent of lung abnormalities. The progressing tendency of imaging pattern and CT score over time were analyzed.
Length of stay prediction model construction
In the LOS prediction model construction, the 1761 CT scans were considered as independent sample and randomly divided into training and testing set with a ratio of 7:3 at patient-level. A combined model was built using imaging reading features, AI features and the corresponding baseline clinical features. To reduce possible variability and provide more stable estimates, feature selection was performed in the training set. During the procedure of feature selection, each feature was normalized by z-score standardized translation. Then, the Least Absolute Shrinkage and Selection Operator (LASSO) was used for feature selection. The selected features were fed into gradient boosting decision tree (GBDT). GBDT is one of the most popular machine learning approach for various tasks in recent years., BGDT is an ensemble learning model with decision trees as base models, which numerical optimized by combing steepest-descent minimization and stage-wise additive expansions in function space. In the GBDT, all decision trees are trained sequentially with the goal of minimizing the residual between the trees-based predictions and the observations. Besides, GBDT model is a stage-wise additive model with the outputs of all decision trees as sequential inputs. We used grid search and stratified five-fold cross validation to select the best parameters for the machine learning models. The best parameters of the GBDT model were found as follows: the number of decision trees was 100, the maximal depth was 3, and the minimum sample split number was 2. Furthermore, to demonstrate the effectiveness of combining imaging features with clinical characteristics, ablation studies on the clinical features, imaging reading features, AI features, and combined features of AI and image reading features were constructed with the same training scheme mentioned above.
Finally, we applied the models from the training set to the testing data. The same features from the training scheme were obtained but the feature values were replaced with those calculated from the testing set, and then normalized with z-score standardized translation using the mean and standard deviation (SD) values derived from the training set.
Continuous variables (e.g., age, gene, blood test indicators and lung infection percentage) are presented as the mean ± SD, while categorical variables were described in the number of cases (proportion). We used Student's t-test or Chi-square test to compare the differences between different admission stages. The model performance was evaluated with mean absolute error (MAE), mean squared error (MSE), R2 coefficient and Pearson correlation coefficient. Statistical analyses were performed with Python version 3.7.0, an open source language administered by the Python Software Foundation. The significance threshold was set as a P < 0.05.
All model constructions were performed and evaluated on a research platform, namely uAI Research Portal (Shanghai United Imaging Intelligence, Co., Ltd.), including image segmentation, feature extraction and selection, model building, and evaluation.
| Results|| |
In this study, 504 patients were involved and 448 patients with confirmed COVID-19 pneumonia were evaluated. As shown in [Table 1], a total of 448 patients underwent the baseline CT examination, and the maximum CT scan number of six was included in this study. A total of 1761 CT scans were finally obtained from the follow-up CT examinations. The range of the hospital admission time was between 0 and 41 days with an overall average time of 22.5 ± 7.0 days. The average interval from the first, second, third, fourth, fifth, and sixth CT scans to the hospital admission time was 1.9 ± 3.2, 5.2 ± 3.6, 9.2 ± 4.3, 12.9 ± 4.2, 16.7 ± 4.6, and 20.7 ± 4.5 days, respectively. Based on the average interval time of CT scan to the initial hospital admission time, we divided the time axis into five periods: 0–3 days (early stage), 4–7 days (progressive stage), 8–11 days (peak stage), 12–15 days (early absorption stage), and ≥16 days (absorption stage).
|Table 1: The number of patients and mean±standard deviation of length of stay in the first and follow-up chest computed tomography examinations|
Click here to view
Statistical results of demographic, disease, and clinical examination characteristics are shown in [Table 2], of which only the initial admission information were employed. In the 448 enrolled patients, the average age is 53.55-year-old, and the male patients accounts for 39.5%. 38.6% of COVID-19 patients have complications with the top two common diseases being diabetes and hypertension. Regarding the clinical symptoms, cough (58.6%), fever (47.1%), and dyspnea (oxygen uptake, 44.3%) are the main clinical symptoms. 35.2% of patients received all recommended doses of the COVID-19 vaccine before the initial PCR positive results and were fully vaccinated over 14 days.
Dynamic changes of different computed tomography patterns
The longitudinal changes in chest CT findings are illustrated in [Table 3] and [Figure 1]. For the lesion-based analysis (e.g., pneumonia infected percentage and number), the progression changes significantly in the earlier stages (0–3 days to 4–7 days), and then changes tends to be statistically subtle [[Table 3] and [Figure 1], P > 0.05]. For the lung field intensity, the healthy lung parenchyma ([−1000, −570] HU) increases significantly over time, while the HU ranges from −470 to −70 shows a significant increase in the early stage followed by a continuous decrease. The CT intensity ≥ −70 HU is gradually recovered to the initial severity-level of symptoms as no significant differences was found between the values of 0–3 days and ≥16 days. For the CT image reading features, ground-glass opacity (GGO), consolidation and crazy paving pattern, air bronchograms and fibrosis strands are the most main CT findings with initial infected ratio of 76,0%, 64.0%, 50.9%, 47.0% and 31.5%, respectively. As shown in the [Figure 1]a, the progression changes significantly in the earlier stages (0–3 days–4–7 days), and then recovered gradually. The pulmonary fibrosis with initial infected ratio of 1.1% shows a significant increase in the early stage followed by a continuous subtle increase. What's more, we further analyzed the top-3 most frequently occurring features (i.e., GGO, consolidation and crazy paving pattern). During the entire course of illness, the superimposed of the three features simultaneously was the main pattern, and then reached the highest on 4–7 days followed by a gradual decrease [Figure 1]b. As depicted in [Figure 2], the bilateral lower lobes, especially the right lower lobe, presents more severe.
|Figure 1: Dynamic changes of CT morphological characteristics on different LOS stage (day). (a) Line chart of image reading features on LOS; (b) The main CT patterns in COVID-19 patients over time, GGO: ground-glass opacity. CT = Computed tomography, LOS = Length of stay|
Click here to view
|Figure 2: Longitudinal changes of infection-related features with LOS (day). LOS = Length of stay|
Click here to view
Feature selection and prediction model construction
For the combined LOS prediction model construction, 37 clinical features, 5 imaging reading feature and 43 AI features of each COVID-19 patient were selected by LASSO. As shown in [Figure 3], the features with nonzero coefficients were illustrated. Feature selected results showed that hospital stay time, oxygen uptake, diarrhea, N gen, immunoglobulin G, density/mass of lesions, infected numbers as well as infected surface distribution are the top related factors.
|Figure 3: The features selected by LASSO feature selection. LASSO = Least Absolute Shrinkage and Selection Operator|
Click here to view
As shown in [Table 4], combining the clinical, imaging reading and AI features to build the LOS prediction model (i.e., AI + Image Reading + Clinical model) achieved the highest R2 of 0.854, Pearson correlation coefficient of 0.939, and lowest MAE of 2.405, and MSE of 9.176 on the training set. And the R2, Pearson correlation coefficient, MAE and MSE is 0.463, 0.696, 4.426 and 34.728 on the testing set. The AI + Image Reading model achieved a MAE of 3.143/4.903, MSE of 15.614/38.096, R2 of 0.751/0.411 and Pearson correlation coefficient of 0.884/0.654 on the training and testing sets, respectively. The AI feature model achieved a MAE of 3.252/4.966, MSE of 16.660/40.052, R2 of 0.734/0.381 and Pearson correlation coefficient of 0.871/0.631 on the training and testing sets, respectively. The MAE, MSE, R2 and Pearson correlation coefficient of the Imaging Reading Feature model is 4.838/4.798, 34.415/38.854, 0.451/0.399 and 0.672/0.644 on the training and testing sets, respectively. For the Clinical Feature model, the MAE, MSE, R2 and Pearson correlation coefficient is 2.419/4.597, 9.509/37.457, 0.848/0.421 and 0.937/0.673 on the training and testing sets, respectively.
| Discussion|| |
In this study, we aim to build a predictive model that predicts COVID-19 patients' LOS by analysis of dynamic CT changes by combining the radiologists, AI features and clinical features. We studied 448 patients with Delta variant of SARS-CoV-2 and a total of 1761 CT scans from the follow-up CT examinations. These patients had multiple chest CT scans at a maximum six different time points which provided reliable data on the dynamic radiologic pattern. After the feature selection, the “AI + Image Reading + Clinical” model has the best performance on the training and testing set.
The clinical and dynamic changes of the image manifestations are particularly important to judge the changes in severity, adjust the treatment plan, and infer the LOS with COVID-19 patient.,, A large number of literatures have reported the dynamic imaging changes of COVID-19,,,,,, but there are few studies on the Delta variant of SARS-COV-2. In these studies, the COVID-19 patients with Delta variant of SARS-CoV-2 were more mild symptoms and the typical imaging findings are similar to those of COVID-19, which are GGO in the subpleural area of both lower lungs.,,
In our study, manual reading plays an important role in this study. As illustrated in [Figure 1], GGO is the primary manifestation in the CT images of patients with COVID-19 and develops dynamically. Its proportion peaks from day 4 to day 7, and after minor diminution, the proportion of GGO reincreases from day 12 to day 15. In the early stages, the infection is still minor and the primary manifestation is scattered GGOs in both lungs. With the development of infection, consolidation appears more and the proportion of GGOs decreases. However, with the treatment takes effect, consolidation absorbs and GGO reappears, which means the infection is getting better. In some patients, there isn't any positive CT primary manifestation regardless the positive result of PCR tests. GGOs don't appear until 4–7 days after admission, which indicates that CT manifestations could appear later than PCR test.
It is obvious that manual reading usually requires heavy workload and longer time, and it can only generally describe the different characteristics of a single patient instead of providing specific quantifications, which helps track the severity of the patient's condition at different times.
| Conclusion|| |
In this study, the GBDT model was used to predict the patients' LOS. Results show that the model based on the combined clinical, imaging reading and AI features obtained the best performance with a coefficient of determination R2 of 46.3% and a MAE of 4.43 days on the testing set. Most of current LOS prediction studies are predominantly focused on assessing the categorical outcomes, such as dividing the LOS into short-stay, medium-stay, and long-stay according to the length of hospitalization of patients. However, it has been demonstrated that the patients' LOS is a right skewed distribution,, which indicates that the dataset is severely unbalanced, as there are only a few cases with long LOS. This imbalance may cause that cases with long LOS are treated as outliers by the model. Comparing to the classification model, a regression model is more suitable for this task. Caetano et al. applied six regression models, including decision trees, random forests, multiple regression, artificial neural network (ANN) ensemble, support vector machine (SVM) and taking average prediction, to compare the predictive performance of LOS. Results suggest that the tree model (random forest) achieved the best results. Mahboub et al. used 2017 COVID-19 patients to build a decision tree model to predict the COVID-19 LOS based on clinical information. The authors reported a R2 of 49.8% and MAE of 2.85 days. While these results show a good prediction capacity, this model is constructed based on 2017 COVID-19 patients collected in 2020, which includes a large proportion of moderate and severe patients. This is different with the current situation where COVID-19 patients is mainly mild or asymptomatic. Therefore, our constructed model is more suitable for the current task of predicting COVID-19 patients with mild or asymptomatic status.
There are several limitations of our study that merit consideration. First, this retrospective analysis is a single site study with a modest sample size, which may lead to suboptimal statistical power when considering subgroup analysis (such as the period of absorption stage). A large sample size will be needed for further study. Second, there is a lack of external testing set to further validate the results of this study, we will include external independent testing set to further validate the results of this study in the future. Third, we managed to obtain the baseline clinical features and used them in the experiments. The clinical features in follow-up scans were not complete enough and thus were not used in this study. Fourth, COVID-19 may affect the patients' myocardium, gastrointestinal tract, nervous system and other organs. We would pay more concerns on these topics in next stage. Finally, since SARS-CoV-2 mutates rapidly, the COVID-19 investigated in this study did not include the latest mutations.
Financial support and sponsorship
This study was supported in part by the 66th Batch of China Postdoctoral Science Foundation Projects (2019M661805) and a Research Grant of Key Project supported by Medical Science and Technology Development Foundation, Nanjing Department of Health (YKK18062), Jiangsu Province, China, the Fundamental Research Funds for the Central Universities (021414380462,021414380484), and the National Science and Technology Innovation 2030-Major Project (2021ZD0111103).
Conflicts of interest
Y.W. and F.S. are employee of Shanghai United Imaging Intelligence Co., Ltd. The company has no role in performing the surveillance and interpreting the data. All other authors have no conflict of interest.
| References|| |
Rodriguez-Morales AJ, Cardona-Ospina JA, Gutiérrez-Ocampo E, Villamizar-Peña R, Holguin-Rivera Y, Escalera-Antezana JP, et al.
Clinical, laboratory and imaging features of COVID-19: A systematic review and meta-analysis. Travel Med Infect Dis 2020;34:101623.
Yang J, Zheng Y, Gou X, Pu K, Chen Z, Guo Q, et al.
Prevalence of comorbidities and its effects in patients infected with SARS-CoV-2: A systematic review and meta-analysis. Int J Infect Dis 2020;94:91-5.
Shi F, Xia L, Shan F, Song B, Wu D, Wei Y, et al.
Large-scale screening to distinguish between COVID-19 and community-acquired pneumonia using infection size-aware classification. Phys Med Biol 2021;66:065031.
Shi F, Wei Y, Xia L, Shan F, Mo Z, Yan F, et al.
Lung volume reduction and infection localization revealed in Big data CT imaging of COVID-19. Int J Infect Dis 2021;102:316-8.
Lin SF, Lin HA, Chuang HC, Tsai HW, Kuo N, Chen SC, et al.
Fever, tachypnea, and monocyte distribution width predicts length of stay for patients with COVID-19: A pioneer study. J Pers Med 2022;12:449.
Shan F, Gao Y, Wang J, Shi W, Shi N, Han M, et al.
Abnormal lung quantification in chest CT images of COVID-19 patients with deep learning and its application to severity prediction. Med Phys 2021;48:1633-45.
Zhou Y, Ren H, Wang S, You F, Feng L, Wang M, et al.
The evolution of chest CT findings from admission to follow-up in 30 moderate to severe adult patients with COVID-19 pneumonia. Chin J Acad Radiol 2021;4:71-7.
Friedman JH. Greedy function approximation: A gradient boosting machine. Ann Stat 2001;29:1189-232.
Gutierrez JM, Volkovs M, Poutanen T, Watson T, Rosella LC. Risk stratification for Covid-19 hospitalization: a multivariable model based on gradient-boosting decision trees. CMAJ Open 2021;9:E1223-E1231.
Cheng J, Chen XH. Research on travel time prediction model of freeway based on gradient boosting decision tree. J Southeast Univ (Engl Ed) 2019;7466-80.
Kwee TC, Kwee RM. Chest CT in COVID-19: What the radiologist needs to know. Radiographics 2020;40:1848-65.
Rubin GD, Ryerson CJ, Haramati LB, Sverzellati N, Kanne JP, Raoof S, et al.
The role of chest imaging in patient management during the COVID-19 pandemic: A Multinational Consensus Statement from the Fleischner Society. Chest 2020;158:106-16.
Li J, Zhao G, Tao Y, Zhai P, Chen H, He H, et al.
Multi-task contrastive learning for automatic CT and X-ray diagnosis of COVID-19. Pattern Recognit 2021;114:107848.
Ozturk T, Talo M, Yildirim EA, Baloglu UB, Yildirim O, Rajendra Acharya U. Automated detection of COVID-19 cases using deep neural networks with X-ray images. Comput Biol Med 2020;121:103792.
Di D, Shi F, Yan F, Xia L, Mo Z, Ding Z, et al.
Hypergraph learning for identification of COVID-19 with CT imaging. Med Image Anal 2021;68:101910.
Wang S, Kang B, Ma J, Zeng X, Xiao M, Guo J, et al.
A deep learning algorithm using CT images to screen for Corona virus disease (COVID-19). Eur Radiol 2021;31:6096-104.
Rangarajan AK, Ramachandran HK. A preliminary analysis of AI based smartphone application for diagnosis of COVID-19 using chest X-ray images. Expert Syst Appl 2021;183:115401.
Ucar F, Korkmaz D. COVIDiagnosis-Net: Deep Bayes-SqueezeNet based diagnosis of the coronavirus disease 2019 (COVID-19) from X-ray images. Med Hypotheses 2020;140:109761.
Wu X, Chen C, Zhong M, Wang J, Shi J. COVID-AL: The diagnosis of COVID-19 with deep active learning. Med Image Anal 2021;68:101913.
Cheng QR, Fan MX, Hao J, Hu XC, Ge XH, Hu ZL, et al.
Chest CT features of children infected by B.1.617.2 (Delta) variant of COVID-19. World J Pediatr 2022;18:37-42.
Yoon SH, Lee JH, Kim BN. Chest CT findings in hospitalized patients with SARSCoV-2: Delta versus omicron variants. Radiology 2022;305:E66.
Chohan F, Ishak A, Alderette T, Rad P, Michel G. Clinical presentation of a COVID-19 Delta variant patient: Case report and literature review. Cureus 2021;13:e18603.
Harerimana G, Kim JW, Jang B. A deep attention model to forecast the Length Of Stay and the in-hospital mortality right on admission from ICD codes and demographic data. J Biomed Inform 2021;118:103778.
Etu EE, Monplaisir L, Arslanturk S, Masund S, Aguwa C, Markevych I. Prediction of length of stay in the emergency department for COVID-19 patients: A Machine Learning Approach. IEEE Access 2022;10:42243-51.
Caetano N, Laureano R, Cortez P. A data-driven approach to predict hospital length of stay a Portuguese case study. ACM 2014;407-14.
Mahboub B, Bataineh MT, Alshraideh H, Hamoudi R, Salameh L, Shamayleh A. Prediction of COVID-19 hospital length of stay and risk of death using artificial intelligence-based modeling. Front Med (Lausanne) 2021;8:592336.
[Figure 1], [Figure 2], [Figure 3]
[Table 1], [Table 2], [Table 3], [Table 4]