Semi-quantitative software evaluation of COVID-19 CT examinations—correlation with clinical parameters

Introduction

Since December 2019, the world has been facing a novel coronavirus variant, namely severe acute respiratory syndrome coronavirus type 2 (SARS-CoV-2) (1-4). In the early stages of the pandemic, detection and proof of infection were of greatest importance and were the focus for chest x-rays and computed tomography (CT) scans (4). However, during different stages of the pandemic, the focus of imaging shifted, and increasing attention has been given to monitoring lung complications and long-term changes, as well as to assessing prognosis and planning therapy (5,6).

Early detection and early assessment of disease severity are associated with reduced mortality in coronavirus disease 2019 (COVID-19) patients (7). Up to 15% of COVID-19 patients develop acute respiratory distress syndrome (ARDS) (8). Prediction of severe disease progression is relevant and of crucial importance, as it allows timely interventions and intensive care capacity planning that may improve the prognosis of patients (7,9). A better estimate of the need for mechanical ventilation or extracorporeal membrane oxygenation (ECMO) could become possible. In this context, chest CT not only allowed a positive diagnosis of COVID-19 but also enabled an overall assessment of the severity of the disease (7). Lung involvement can be quantified on chest CT with some classification schemes and prognostic value (10,11). A correlation between the need for oxygenation support and intubation with compromised lung volume on chest CT scans in COVID-19 patients has been reported (12,13). Furthermore, the relationship between the severity of lung involvement on chest CT scans in COVID-19 positive patients with clinical and laboratory data is known (12,13). Quantitative lung CT analysis has acquired an increasingly relevant role in the clinical evaluation and management of varying diseases affecting the lung (14,15). Here radiological evidence of extensive parenchymal affection is described to be related to severe disease and worse outcome (16). However, detailed evaluation in clinical routine is time-consuming and depends on the experience of the individual radiologist (16). In this context, software approaches can support radiologists by providing them with dedicated information about the distribution of opacities in affected lung lobes and their individual appearance in correlation with clinical parameters. Different severity scores related to disease progression and prognostic scores related to CT data have been introduced based on the results of artificial intelligence (AI) algorithms (17-20). An established deep-learning software approach was the pneumonia analysis software, which automatically segments lung volume, identifies areas of COVID-19 pneumonia, and quantifies affected and non-affected areas of the lung (5,6,21-23).

Against this background, in this retrospective single-center study, we aimed to quantify CT parameters of affected lungs in COVID-19 patients for severity assessment in correlation with clinical and laboratory parameters. Furthermore, this study also aimed at an analysis of the likelihood of staying in the intensive care unit (ICU), the need for intubation, or even death. We present this article in accordance with the STROBE reporting checklist (available at https://jtd.amegroups.com/article/view/10.21037/jtd-2025-655/rc).

Methods

Compliance with ethical standards

This study was conducted in accordance with the Declaration of Helsinki and its subsequent amendments. The study was approved by ethics board of Ulm University (No. 438/20). The requirement for written informed consent was waived due to the retrospective design of this study.

Study population

This single-center retrospective study included 66 consecutive patients with lung CT datasets acquired between 12/2020 and 05/2021. The case selection criteria are detailed in Figure 1. All patients were identified by database querying of the Department of Diagnostic and Interventional Radiology at Ulm University.

Figure 1 Patient recruitment. COVID-19, coronavirus disease 2019; CT, computed tomography; PCR, polymerase chain reaction.

Exclusion criteria were patients with no CT examination within 3 days before or after positive polymerase chain reaction (PCR) test and patients under 18 years of age.

CT examinations

All CT examinations were performed using multi-slice CT scanners [SIEMENS SOMATOM Edge Plus (56/66, 84.8%), SIEMENS SOMATOM Force (10/66, 15.2%)]. The field of view was adjusted for each patient to include the entire chest wall and both lungs. Intravenous contrast agent was administered in 25/66 (38%) patients. In all patients, a single spiral acquisition was performed from the lung apex to lung base during one breath-hold at suspended end-inspiratory volume. All examinations were performed in supine position.

Software solution

Image analysis was performed using the syngo.via CT Pneumonia Analysis Software (version 1.0.4.2, Siemens Healthineers, Erlangen, Germany). The software uses Digital Imaging and Communication in Medicine (DICOM) images. The algorithm automatically delineates airspace opacities using a convolutional neural network trained on data manually labeled by clinical experts. The calculation of the opacity scores is described in detail by Bernheim et al. (23).

Further, a three-dimensional (3D) neural network was trained to separate between COVID-19 and non-COVID-19. The model receives as input a two-channel 3D tensor, where the first channel contains the CT Hounsfield units (HU) and the second channel contains the probability map of the previous opacity classifier. The model was trained end-to-end as a classification system using binary cross-entropy. It also used probabilistic sampling of the training data to adjust for label imbalance in the training dataset. Details can be found in Mortani Barbosa et al. (21). The model of initially trained software is illustrated in Figure 2. All technical details of the software parts, i.e., the development of the algorithms, the different post-processing steps, and the evaluated parameters, have been described in detail before (5,6,21-23).

Figure 2 Network training for pneumonia analysis software. COVID-19, coronavirus disease 2019; CT, computed tomography; DL, deep learning.

The calculations were based on two measurements: a global analysis was performed first, followed by a lobe-specific analysis (6). Definitions and details of the measurements are given in Table 1.

All volumes (whole lung, individual lung lobes, and affected tissue) are expressed as percentages and in milliliters. The mean densities of the whole lungs, lung lobes, and areas suspected of being affected by COVID-19 are reported in HU. An HU threshold of −200 HU was applied to identify high opacities (13). A lung lobe was classified as “affected” if the algorithm detected high opacity abnormalities in a defined lung part.

Manual corrections were made after the automatic analysis was completed. Lung lobes were examined and adjusted accordingly. Extensive corrections were required in patients with soft tissue emphysema, where air was confused with lung tissue, and in patients with pleural effusion. Discrepancies between vessel markings and consolidations due to contrast agent were also corrected. Motion artifacts and foreign bodies such as stents and clips were identified, and artifacts appearing as ground-glass opacities or consolidations were removed from the analysis as far as possible. Examples were given in Figures 3,4.

Figure 3 The image depicts the assessment of a 28-year-old male patient with a CT-value of 21.8 HU. The infection was diagnosed during a check-up for B-NHL as the underlying tumor disease. The patient presented no symptoms, was hospitalized for 22 days, and did not require intensive care. Morphologically, atypical pneumonic infiltrates consistent with the condition were observed, with an overall mild to moderate manifestation. The probability of COVID-19 infection was 100%. B-NHL, B-cell non-Hodgkin lymphoma; COVID-19, coronavirus disease 2019; CT, computed tomography; HU, Hounsfield units.

Figure 4 The image presents an 85-year-old male patient with a CT-value of 32.2 HU. The patient exhibited infection symptoms for one day, including exposure to COVID-19, fatigue, altered taste, nausea, and vomiting. His medical history included a tumor condition (pancreatic carcinoma), arterial hypertension, obesity, and a history of smoking. The patient was hospitalized in a general ward for 12 days and did not require intensive care. Morphologically, bilateral pulmonary artery embolism, predominantly on the right side, and bipulmonary atypical infiltrates consistent with COVID-19 were observed. The probability of COVID-19 infection was 100%. COVID-19, coronavirus disease 2019; CT, computed tomography; HU, Hounsfield units.

Image and post-processing analyses

All scans were viewed with standard mediastinal windows (level 35 HU; width 450 HU) and lung windows (level −700 HU; width 1,500 HU). Furthermore, CT quality was classified as perfect, good, moderate, or inadequate. All examinations were reviewed by two independent readers (N.T., M.B.). Additionally, a subgroup classification was performed regarding technical post-processing criteria. The criteria are shown in Table 2.

Virological standard of reference

In all patients with COVID-19, the diagnosis was confirmed by reverse transcriptase polymerase chain reaction (RT-PCR) assay. In detail, viral infection was detected and verified in 42/66 patients by throat swab, in 16/66 patients by nasopharyngeal swab, in 1/66 patients by bronchial lavage, in 1/66 patients by nasopharyngeal secretions, and in 1/66 patients by tracheal secretion. The detection method was not specified for 5 out of 66 patients. All tests were followed by an RT-PCR assay to confirm the diagnosis.

Study design

Patient symptoms were evaluated, including typical infection symptoms of the bronchial system, including fever, fatigue, cough and dyspnea. Furthermore, COVID-19-specific symptoms like a disturbed sense of smell and diarrhea were evaluated. COVID-19 symptoms and laboratory data were retrospectively also collected from the electronic medical records platform.

C-reactive protein (CRP), creatine kinase (CK), troponin T, aspartate transaminase (AST), alanine transaminase (ALT), lactate dehydrogenase (LDH), white blood cell count (LEU), D-dimer, ferritin, procalcitonin, interleukin 6 and lactate were collected at the time point of CT performance (±24 h) and were analyzed to examine a possible relation with automated quantification of abnormalities associated with COVID-19.

For sub-cohort analysis, patients were divided into survivors and deceased patients, as well as into patients who needed intensive care medicine or intubation. Furthermore, it was evaluated whether the patient required high flow oxygen or non-invasive ventilation (NIV) therapy.

Statistical analysis

Statistical analysis was performed using SPSS software (version 27.0; IBM Corp., Armonk, NY, USA). All results are expressed as mean and standard deviation. The Kolmogorow-Smirnov test was used for normality testing, including Lilliefors significance correction. Mann-Whitney U tests and paired t-tests were performed to compare different subgroups. For subgroup analysis, Chi-squared tests were performed to confirm the frequency and distribution within each group. Post-hoc tests with Bonferroni adjustments were conducted to adjust for multiple comparisons. For lung quantification parameters, paired t-tests were performed for mean HU of opacity and Mann-Whitney U test was used for all other parameters to compare subgroups with and without applied contrast agent.

Correlations were tested using Pearson correlation analyses. The level of statistical significance was set at P<0.05 (two-sided).

Results

Cohort characteristics

This study included 66 consecutive patients (31 females, mean age 64.6±16.2 years) with lung CT datasets from 12/2020 to 05/2021. The mean time from CT examination to virological proof was 0.33±0.67 days. The mean cycle threshold value from RT-PCR was 26.15±5.81. Overall, the mean cycle threshold value was lower in deceased patients, but not significantly different (23.25±5.03 vs. 26.82±5.87, P=0.55). Overall, the mean length of stay in the ICU was significantly longer in deceased patients (9±9 vs. 1±4 days, P=0.009).

Characteristic symptoms and underlying diseases are provided in Table 3. In 23/66 cases high flow oxygen therapy or NIV treatment was performed. Oxygen support by nasal cannula or Ohio mask was used in 47/66 cases. Furthermore, ECMO therapy was required in 2/66 cases.

Image analysis/image quality and post-processing

Overall, 26/66 (39.4%) examinations were of perfect quality, 33/66 (50.0%) examinations were of good quality, and 7/66 (10.6%) examinations were of moderate quality. Inadequate examinations were excluded. Corrections were classified as minor and major: major corrections were required in 26/66 cases (39.4%) and minor corrections in 40/66 cases (60.6%).

Mean time to upload the patient data to the software was 0.29±0.13 min, and the mean time from starting the software with patient data to modifications was 2.72±0.51 min. Time to save the results took 1.36±0.50 min. Mean post-processing time to complete the full analysis (includes the entire editing with individual adjustments) was 27.32±18.02 min.

Patient outcome/intensive care medicine

Of 66 patients, 12 died (18.2%) and 10 (15.2%) had to be intubated. Admission to the ICU was required in 27/66 (40.9%) cases, with time on the ICU amounting to 5.7±10.8 days on average (range, 1–65 days). The mean time of mechanical ventilation was 3.0±8.4 days (range, 0–52 days). Mean time of hospital stay was 16.1±13.7 days (range, 0–64 days).

Quantitative CT analysis

The mean estimated probability of COVID-19 in the entire cohort was 0.81±0.36. The lung severity score (LSS), reflecting the severity of all lung lobes, was 7±4.7. The total lung volume was 3,903.65±1,185.67 mL and the mean opacified volume was 866.52±829.29 mL. A general overview is given in Table 4, including a subgroup overview for cases with applied intravenous contrast agent and without. Furthermore, a summary of the assessed pulmonary parameters in subgroups classified according to different levels of need for intensive care treatment is shown in Table 5. A subgroup analysis for patients with and without applied contrast agent is given in Tables S1,S2.

The mean probability of COVID-19 in deceased patients was not significantly different (P=0.22) compared to patients that survived, but volume of opacity and high opacity were significantly higher (P=0.007 and 0.04) in deceased patients (Figures 5,6).

Figure 5 Comparison of mean volume of opacities and volume of high opacities in subgroups of deceased patients and patient that recovered.

Figure 6 Comparison of mean percentage of opacities and percentage of high opacities in subgroups of deceased patients and patient that recovered.

In patients who need intensive care treatment with extended lung opacities, the volume of opacity (P<0.001), percentage of opacity (PO) (P<0.001), volume of high opacity (P<0.001), and percentage of high opacity (POH) (P<0.001) were significantly higher. Lobe-specific analysis of COVID-19 pneumonia is provided in Table S3.

Volume of opacities correlates well with days of intubation (r=0.554, P<0.001) and with days with intensive care medicine (r=0.554, P<0.001), however not with overall days in hospital (r=0.075, P=0.55).

Quantification parameters volume of opacity, PO, volume of high opacity and POH showed moderate to strong correlations to days of intubation and days with intensive care medicine (each <0.001), however, not to overall days in hospital (P>0.05). Details were given in Table 6, overall correlation results are provided in Table S4.

Laboratory parameters

An overview of the evaluated laboratory parameters is given in Table 7. In deceased patients, increased D-dimer (P=0.02), procalcitonin (P=0.01), lactate (P=0.009), CRP (P=0.003), and LDH (P=0.002) were registered compared to patients that survived.

Similar to deceased patients, a significant increase in infection parameters was observed in patients requiring intensive care treatment: LDH (P=0.002), CRP (P=0.001), procalcitonin (P<0.001), and interleukin 6 (P=0.047). No significant differences were found among other laboratory constellations, underlying diseases, and clinical symptoms.

Volume of opacity (mL) correlates moderately positively with inflammation blood markers: with CRP (r=0.478, P<0.001) and with procalcitonin (r=293, P=0.02).

Volume of high opacity eves correlates strongly positive with CRP (r=0.619, P<0.001), with procalcitonin (r=44, P=0.02) and even with interleukin-6 (r=0.609, P=0.002). An overview of the most relevant correlations is given in Table 8, a full analysis is provided in Table S2.

Discussion

During the COVID-19 pandemic, lung CT imaging played an important role, first in screening and later in assessing and monitoring the disease burden (24). In our study, in addition to visual assessment, an AI-augmented approach for quantitative image analysis provided precise information on lung involvement and showed associations with clinical information and intensive care treatment. Furthermore, AI-augmented approaches are technically feasible and have been continuously investigated in different settings since the onset of the pandemic (25-28). In this context, our study demonstrated the clinical usefulness of an AI-augmented software to semi-automatically quantify COVID-19-related lung opacities in chest CT in relation to intensive care medicine, clinical, and laboratory data.

In our cohort, lower cycle threshold values correlated with higher disease severity and were associated with progression to more severe disease courses, which is in accordance with the current literature (29-31). The cycle threshold values as predictors of disease severity were related to higher in-hospital mortality when compared to patients with lower viral loads (32). Ashrafi et al. showed that patients with renal insufficiency with cycle threshold values ≤20 had a higher rate of intensive care admission than those with cycle threshold values >20 (32). Jemmieh et al. showed that the median cycle threshold value was slightly lower in deceased patients (23.25±5.03) than in survivors (26.80±5.82) (33).

Looking more closely at the need for intensive care treatment, our results are in line with current literature highlighting this aspect during the pandemic: Kurzeder et al. showed in 289 patients that 29% required nasal oxygen administration, 28% were admitted to the ICU, and 15% died (34). In our cohort, the mortality rate was slightly higher at 18.2%, which might be partly due to the fact that our hospital is a large academic facility providing maximum care. This suggests that more severe cases are sometimes referred from surrounding hospitals. Our results also reflect that higher age was significantly associated with mortality, which is in accordance with this previous study (34).

Mader et al. reported in a group of 50 patients that 16% of the patients died, also performing analyses on data from a university center (13). In their cohort, 48% (24/50) of patients required intensive care admission, with a mean length of 15.62±13.46 days (13). In our cohort, mean time on ICU was slightly shorter at 5.72±10.87 days, which may be due to different intermediate care options at our site. The mean time of stay at hospital in our cohort was 16.12±13.73 days, which is similar to their results with a mean stay of 17.22 days (13). Arru et al. showed in their multi-center study both a reduction in mortality (10–12%) and a reduction in the need for ICU admission (19–32%), possibly due to the peculiarities of the different sites (35).

The crucial importance of the percentage and volume of pulmonary opacities is that they are superior features in predicting patient outcomes (6,12,36). Our results, as well as another study, highlight that quantitative CT measurements may predict adverse outcomes in COVID-19 patients (28). The total volume of opacities showed high performance in assessing the severity of COVID-19 and was higher in patients who died, which is in line with previous studies by Pang et al. and Sun et al. (28,37). Our finding that areas of high opacity were higher in deceased patients is also in line with previous work (9,38).

Similar to our results, Mader et al. showed that higher opacity scores, higher PO, and higher PHO were registered in patients with the need for intensive care treatment, thus most likely reflecting a higher disease burden (13). Our results are slightly inferior to earlier findings regarding disease distribution, with a mean opacity percentage of 23.54 in both lungs compared to 28.40 in the cohort of Mader et al. (13). Correspondingly, the POH for both lungs was slightly lower with 5.69 as compared to 10.46 in their cohort (13).

In their bicentric study, Homayounieh et al. showed that the best predictors of the final outcome were the volume of all attenuation opacities and the percentage of high attenuation, similar to prior studies (12,39). Overall, this again highlights the predictive value of quantitative features and may underline the robust performance of the software (39). Most common clinical symptoms were dyspnea and fever, similar to current literature (13). Interestingly, in our cohort, there were no significant clusters of symptoms or significant predisposing underlying diseases. This is best explained by the fact that the number of individual cases in the subgroups is low.

Compared to previous studies, the correlations of the parameters we quantified showed two special features: in particular, the parameters correlate with days of intubation and days with intensive care medicine, however not to overall days in hospital. In previous literature there were already lower correlations with the total length of stay in hospital, but our data suggests that the total length of stay may be due to other aspects such as underlying illnesses, or through the infrastructure of the hospital or the local aftercare connections (13).

The second point that differs from previous studies is the strong correlation of the quantified subscores [PO, vertical horizontal opacity (VHO) and PHO] with clinical parameters. The software we use goes beyond simply determining the affected volume.

Evaluating the technical performance of the software, all datasets required minor or even major corrections. The mean post-processing time of more than 27 min was longer compared to other studies using the same CT pneumonia analysis approach (6,36). At this point, the mean processing time included the entire editing with individual adjustments in individual cases. This is the most extensive and most time-consuming factor in post-processing, especially for a cohort with more severe cases than for a cohort with only mild cases. Overall, the time required most likely relates considerably to the individual severity of the patients’ illness. In this regard, the majority of our patients required intensive care treatment during the course of the disease.

The approach we use enables semi-quantitative analysis without the additional approach required by radiomics. Radiomics means the concept that medical images are not just visual representations of tissues and organs, but also contain a lot of information about underlying tissue properties and disease patterns. The necessary approach of extracting features from the segmented images, followed by analysis of the extracted features and application of additional AI models, was not implemented.

The software primarily excludes pulmonary vessels, so the application of contrast agent should not have any negative effects on the results. In a detailed analysis, even in our cohort, no significant difference in measured mean HU and measured mean HU in opacities was registered, which matches findings from a previous study (6). This testifies to the accuracy of the software and the accuracy of the annotation performed. Nevertheless, the contrast-media group showed significantly larger volumetric infiltrates, significantly more portions with high opacity, and a significantly higher opacity score. This is likely due to the clinical context, in which patients who were in worse condition during this phase of the coronavirus pandemic were targeted for maximum diagnostics, including assessment of the pulmonary arteries and pulmonary complications. Thus, in our cohort, contrast agent administration was likely significantly more frequent in clinically worse patients.

Our study has several limitations. First, it was a single-center study with a relatively small number of COVID-19 cases. The next step should be a validation in a larger cohort. Second, the retrospective study design does not allow for information on patient follow-up status. Third, a subgroup analysis of the different virus variants and subtypes should be pursued as a next step. Fourth, all patients were examined only once at the peak of the disease, and no long-term analysis was performed. Fifth, as a technical limitation, our analysis only includes examinations from Siemens multi-slice CT scanners, without analyzing images from other manufacturers or comparing with other scanners. This limits the generalizability of our results, as they only represent an analysis of manufacturer-specific image characteristics without examination from other vendors.

Sixth, the software requires human correction and follow-up in all cases. In some cases, this is very minimal, but in others, significant corrections were necessary, which are sometimes time-consuming and require training for human readers.

Conclusions

The pneumonia analysis software may allow semi-quantitative analysis of COVID-19 pneumonia. It can provide information on the extent of opacification on lung CT and, based on this, connections with patient outcome and need for intensive care medicine can be evaluated. Risk assessment is possible using both quantitative CT and clinical parameters.

Acknowledgments

None.

Footnote

Reporting Checklist: The authors have completed the STROBE reporting checklist. Available at https://jtd.amegroups.com/article/view/10.21037/jtd-2025-655/rc

Data Sharing Statement: Available at https://jtd.amegroups.com/article/view/10.21037/jtd-2025-655/dss

Peer Review File: Available at https://jtd.amegroups.com/article/view/10.21037/jtd-2025-655/prf

Funding: This work was funded by “NUM 2.0” as part of the Racoon Project (FKZ: 01KX2121).

Conflicts of Interest: All authors have completed the ICMJE uniform disclosure form (available at https://jtd.amegroups.com/article/view/10.21037/jtd-2025-655/coif). F.F. is currently an employee of Siemens Healthineers, Erlangen, Germany. The other authors have no conflicts of interest to declare.

Ethical Statement: The authors are accountable for all aspects of the work in ensuring that questions related to the accuracy or integrity of any part of the work are appropriately investigated and resolved. This study was conducted in accordance with the Declaration of Helsinki and its subsequent amendments. The study was approved by ethics board of Ulm University (No. 438/20). The requirement for written informed consent was waived due to the retrospective design of this study.

Open Access Statement: This is an Open Access article distributed in accordance with the Creative Commons Attribution-NonCommercial-NoDerivs 4.0 International License (CC BY-NC-ND 4.0), which permits the non-commercial replication and distribution of the article with the strict proviso that no changes or edits are made and the original work is properly cited (including links to both the formal publication through the relevant DOI and the license). See: https://creativecommons.org/licenses/by-nc-nd/4.0/.

References

1. Huang C, Wang Y, Li X, et al. Clinical features of patients infected with 2019 novel coronavirus in Wuhan, China. Lancet 2020;395:497-506. [Crossref] [PubMed]
2. Zhou P, Yang XL, Wang XG, et al. A pneumonia outbreak associated with a new coronavirus of probable bat origin. Nature 2020;579:270-3. [Crossref] [PubMed]
3. Wu F, Zhao S, Yu B, et al. A new coronavirus associated with human respiratory disease in China. Nature 2020;579:265-9. [Crossref] [PubMed]
4. Rubin GD, Ryerson CJ, Haramati LB, et al. The Role of Chest Imaging in Patient Management during the COVID-19 Pandemic: A Multinational Consensus Statement from the Fleischner Society. Radiology 2020;296:172-80. [Crossref] [PubMed]
5. Alarcón-Rodríguez J, Fernández-Velilla M, Ureña-Vacas A, et al. Radiological management and follow-up of post-COVID-19 patients. Radiologia (Engl Ed) 2021;63:258-69. [Crossref] [PubMed]
6. Ebong U, Büttner SM, Schmidt SA, et al. Quantitative Evaluation of COVID-19 Pneumonia CT Using AI Analysis-Feasibility and Differentiation from Other Common Pneumonia Forms. Diagnostics (Basel) 2023;13:2129. [Crossref] [PubMed]
7. Wu G, Zhu Y, Qiu X, et al. Application of clinical and CT imaging features in the evaluation of disease progression in patients with COVID-19. BMC Pulm Med 2023;23:329. [Crossref] [PubMed]
8. Bruneau T, Wack M, Poulet G, et al. Circulating Ubiquitous RNA, A Highly Predictive and Prognostic Biomarker in Hospitalized Coronavirus Disease 2019 (COVID-19) Patients. Clin Infect Dis 2022;75:e410-7. [Crossref] [PubMed]
9. Wong CKH, Au ICH, Lau KTK, et al. Real-world effectiveness of molnupiravir and nirmatrelvir plus ritonavir against mortality, hospitalisation, and in-hospital outcomes among community-dwelling, ambulatory patients with confirmed SARS-CoV-2 infection during the omicron wave in Hong Kong: an observational study. Lancet 2022;400:1213-22. [Crossref] [PubMed]
10. Salahshour F, Mehrabinejad MM, Nassiri Toosi M, et al. Clinical and chest CT features as a predictive tool for COVID-19 clinical progress: introducing a novel semi-quantitative scoring system. Eur Radiol 2021;31:5178-88. [Crossref] [PubMed]
11. Nokiani AA, Shahnazari R, Abbasi MA, et al. COVID-19 Chest CT Quantification: Triage and Prognostic Value in Different Ages. Clin Med Res 2023;21:14-25. [Crossref] [PubMed]
12. Lanza E, Muglia R, Bolengo I, et al. Quantitative chest CT analysis in COVID-19 to predict the need for oxygenation support and intubation. Eur Radiol 2020;30:6770-8. [Crossref] [PubMed]
13. Mader C, Bernatz S, Michalik S, et al. Quantification of COVID-19 Opacities on Chest CT - Evaluation of a Fully Automatic AI-approach to Noninvasively Differentiate Critical Versus Noncritical Patients. Acad Radiol 2021;28:1048-57. [Crossref] [PubMed]
14. Negroni D, Zagaria D, Paladini A, et al. COVID-19 CT Scan Lung Segmentation: How We Do It. J Digit Imaging 2022;35:424-31. [Crossref] [PubMed]
15. Tonelli R, Smit MR, Castaniere I, et al. Quantitative CT-analysis of over aerated lung tissue and correlation with fibrosis extent in patients with idiopathic pulmonary fibrosis. Respir Res 2024;25:359. [Crossref] [PubMed]
16. Barbosa CS, Chaves GWOG, de Oliveira CV, et al. COVID-19 pneumonia in the emergency department: correlation of initial chest CT findings with short-term outcome. Emerg Radiol 2020;27:691-9. [Crossref] [PubMed]
17. Galzin E, Roche L, Vlachomitrou A, et al. Additional value of chest CT AI-based quantification of lung involvement in predicting death and ICU admission for COVID-19 patients. Res Diagn Interv Imaging 2022;4:100018. [Crossref] [PubMed]
18. Li L, Qin L, Xu Z, et al. Using Artificial Intelligence to Detect COVID-19 and Community-acquired Pneumonia Based on Pulmonary CT: Evaluation of the Diagnostic Accuracy. Radiology 2020;296:E65-71. [Crossref] [PubMed]
19. Lessmann N, Sánchez CI, Beenen L, et al. Automated Assessment of COVID-19 Reporting and Data System and Chest CT Severity Scores in Patients Suspected of Having COVID-19 Using Artificial Intelligence. Radiology 2021;298:E18-28. [Crossref] [PubMed]
20. Kwon YJF, Toussie D, Finkelstein M, et al. Combining Initial Radiographs and Clinical Variables Improves Deep Learning Prognostication in Patients with COVID-19 from the Emergency Department. Radiol Artif Intell 2021;3:e200098. [Crossref] [PubMed]
21. Mortani Barbosa EJ Jr, Georgescu B, Chaganti S, et al. Machine learning automatically detects COVID-19 using chest CTs in a large multicenter cohort. Eur Radiol 2021;31:8775-85. [Crossref] [PubMed]
22. Chaganti S, Grenier P, Balachandran A, et al. Automated Quantification of CT Patterns Associated with COVID-19 from Chest CT. Radiol Artif Intell 2020;2:e200048. [Crossref] [PubMed]
23. Bernheim A, Mei X, Huang M, et al. Chest CT Findings in Coronavirus Disease-19 (COVID-19): Relationship to Duration of Infection. Radiology 2020;295:200463. [Crossref] [PubMed]
24. Lu W, Wei J, Xu T, et al. Quantitative CT for detecting COVID‑19 pneumonia in suspected cases. BMC Infect Dis 2021;21:836. [Crossref] [PubMed]
25. Pu J, Leader JK, Bandos A, et al. Automated quantification of COVID-19 severity and progression using chest CT images. Eur Radiol 2021;31:436-46. [Crossref] [PubMed]
26. Chen H, Zeng M, Wang X, et al. A CT-based radiomics nomogram for predicting prognosis of coronavirus disease 2019 (COVID-19) radiomics nomogram predicting COVID-19. Br J Radiol 2021;94:20200634. [Crossref] [PubMed]
27. Homayounieh F, Babaei R, Karimi Mobin H, et al. Computed Tomography Radiomics Can Predict Disease Severity and Outcome in Coronavirus Disease 2019 Pneumonia. J Comput Assist Tomogr 2020;44:640-6. [Crossref] [PubMed]
28. Pang B, Li H, Liu Q, et al. CT Quantification of COVID-19 Pneumonia at Admission Can Predict Progression to Critical Illness: A Retrospective Multicenter Cohort Study. Front Med (Lausanne) 2021;8:689568. [Crossref] [PubMed]
29. Magleby R, Westblade LF, Trzebucki A, et al. Impact of Severe Acute Respiratory Syndrome Coronavirus 2 Viral Load on Risk of Intubation and Mortality Among Hospitalized Patients With Coronavirus Disease 2019. Clin Infect Dis 2021;73:e4197-205. [Crossref] [PubMed]
30. Huang JT, Ran RX, Lv ZH, et al. Chronological Changes of Viral Shedding in Adult Inpatients With COVID-19 in Wuhan, China. Clin Infect Dis 2020;71:2158-66. [Crossref] [PubMed]
31. Phillips MC, Quintero D, Wald-Dickler N, et al. SARS-CoV-2 cycle threshold (Ct) values predict future COVID-19 cases. J Clin Virol 2022;150-151:105153. [Crossref] [PubMed]
32. Ashrafi S, Pourahmad Kisomi P, Maroufizadeh S, et al. The relationship between CT value and clinical outcomes in renal patients with COVID-19. Int Urol Nephrol 2023;55:697-709. [Crossref] [PubMed]
33. Jemmieh K, Tawengi M, Alyaarabi T, et al. No Association Between Ct Value and COVID-19 Severity and Mortality in Qatar. Int J Gen Med 2023;16:5323-31. [Crossref] [PubMed]
34. Kurzeder L, Jörres RA, Unterweger T, et al. A simple risk score for mortality including the PCR Ct value upon admission in patients hospitalized due to COVID-19. Infection 2022;50:1155-63. [Crossref] [PubMed]
35. Arru C, Ebrahimian S, Falaschi Z, et al. Comparison of deep learning, radiomics and subjective assessment of chest CT findings in SARS-CoV-2 pneumonia. Clin Imaging 2021;80:58-66. [Crossref] [PubMed]
36. Gouda W, Yasin R. COVID-19 disease: CT Pneumonia Analysis prototype by using artificial intelligence, predicting the disease severity. Egypt J Radiol Nucl Med 2020;51:1-11.
37. Sun D, Li X, Guo D, et al. CT Quantitative Analysis and Its Relationship with Clinical Features for Assessing the Severity of Patients with COVID-19. Korean J Radiol 2020;21:859-68. [Crossref] [PubMed]
38. Wu D, Gong K, Arru CD, et al. Severity and Consolidation Quantification of COVID-19 From CT Images Using Deep Learning Based on Hybrid Weak Labels. IEEE J Biomed Health Inform 2020;24:3529-38. [Crossref] [PubMed]
39. Homayounieh F, Bezerra Cavalcanti Rockenbach MA, Ebrahimian S, et al. Multicenter Assessment of CT Pneumonia Analysis Prototype for Predicting Disease Severity and Patient Outcome. J Digit Imaging 2021;34:320-9. [Crossref] [PubMed]