A novel deep learning-based quantification of serial chest computed tomography in Coronavirus Disease 2019 (COVID-19)

doi:10.21203/rs.3.rs-38083/v1

Download PDF

Research Article

A novel deep learning-based quantification of serial chest computed tomography in Coronavirus Disease 2019 (COVID-19)

https://doi.org/10.21203/rs.3.rs-38083/v1

This work is licensed under a CC BY 4.0 License

Version 1

posted

You are reading this latest preprint version

Objectives: This study aims to explore and compare a novel deep learning-based quantification with the conventional semi-quantitative computed tomography (CT) scoring for the serial chest CT scans of COVID-19.

Materials and Methods: 95 patients with confirmed COVID-19 and a total of 465 serial chest CT scans were involved, including 61 moderate patients (moderate group, 319 chest CT scans) and 34 severe patients (severe group, 146 chest CT scans). Conventional CT scoring and deep learning-based quantification were performed for all chest CT scans for two study goals: 1. Correlation between these two estimations; 2. Exploring the dynamic patterns using these two estimations between moderate and severe groups.

Results: The Spearman’s correlation coefficient between these two estimation methods was 0.920 (p<0.001). predicted pulmonary involvement (CT score and percent of pulmonary lesions calculated using deep learning-based quantification) increased more rapidly and reached a higher peak on 23^rd days from symptom onset in severe group, which reached a peak on 18^th days in moderate group with faster absorption of the lesions.

Conclusions: The deep learning-based quantification for COVID-19 showed a good correlation with the conventional CT scoring and demonstrated a potential benefit in the estimation of disease severities of COVID-19.

Infectious Diseases

Artificial Intelligence and Machine Learning

X-Ray Computed Tomography

COVID-19

Artificial Intelligence

Coronavirus Disease 2019 (COVID–19) has become a global pandemic since the first report in December 2019 in China ¹. The global number of infections continued to grow to approach 2 million by 16 April 2020 ². Real-time reverse transcription-polymerase chain reaction (RT-PCR) test as the golden diagnostic modality presented a high false-negative rate of nearly one-third, which requires serial tests to avoid missed diagnosis ^3,4. Instead, the chest computed tomography (CT) demonstrated a higher sensitivity of 97% and was increasingly identified as a better screening and monitoring method in the clinical practice ^5,6.

Previous studies revealed the chest CT patterns of COVID–19 with a typical change from early subpleural ground-glass opacity (GGO) to extensive consolidation, which started to be absorbed after at least two weeks from symptom onset ^6–8. In addition, multiple studies confirmed the correlation between higher conventional CT scores and worse prognosis of COVID–19 ^9–11. However, the conventional CT scoring system is semi-quantitative and requires intensive work of the radiologists, which is challenging at the rapid increase of the infected population. On the other hand, to date, there has no ideal tool to automatically quantify viral pneumonia on CT imaging. A deep learning-based module has been set up in our center to establish a more objective and stable evaluation system for the CT estimation of the COVID–19 disease course. This study aims to explore the correlation between the conventional CT scoring system and this novel deep learning-based quantification.

This study was approved by the Ethics of Committees of Union Hospital, Tongji Medical College, Huazhong University of Science and Technology (No. 2020–0026), and followed the 1964 Helsinki Declaration and its later amendments or comparable ethical standards. Only the anonymous data was collected and informed consent for this retrospective study was waived.

Patients and Groups

931 consecutive records for patients with RT-PCR confirmed COVID–19 were reviewed retrospectively for the period from 27th January 2020 to 30th March 2020 in two newly established isolation centers (Western Campus and Zhuankou Fangcang’ Shelter Hospitals) of Union Hospital, Tongji Medical College, Huazhong University of Science and Technology. Considering the potential impact of time from symptom onset on the CT manifestations, only the recovered patients with at least three times of serial chest CT scans were involved ^7,8. Patients with the medical history of pulmonary disease or lung surgery were excluded. Patients with respiratory rate > 30 breaths/min or SpO2 ≤ 93% on room air were classified as severe COVID–19, otherwise as moderate COVID–19 ^12,13. In the end, 95 patients with a total of 465 chest CT scans were involved, including 61 moderate patients (moderate group) and 34 severe patients (severe group).

CT Scan Protocol

The pulmonary CT scans were performed using a single inspiratory phase on a multi-detector CT scanner (Philips Ingenuity Core128, Philips Medical Systems, Best, the Netherlands). The images were obtained during a single breath-hold. The fixed tube voltage was set to 120kVp with an automatic tube current. From the raw data, CT images were reconstructed with a matrix size of 512 × 512 as axial images (thickness of 1.5mm and increment of 1.5mm) in transverse slice orientation with iDose5 iterative reconstruction (Philips Healthcare, Best, Netherlands).

Chest CT Estimation by Radiologists

The major CT demonstrations were described using internationally standard nomenclature defined by the Fleischner Society glossary and peer-reviewed literature on COVID–19, including ground-glass opacity (GGO) and consolidation ^8,14–16. A conventional semi-quantitative scoring system (CT score) was used to estimate the involved pulmonary volume of all these abnormalities ^8,17. There was a score of 0–5 corresponding to the percentage of pulmonary involvement in each lobe as: 0, 0%; 1, <5%; 2, 6%–25%; 3, 26%–49%; 4, 50%–75%; 5, >75%. The scores in five lobes were summed resulting in a total CT score ranging from 0 to 25. Two experienced radiologists (BL and LY, who had 25 and 22 years of experience in thoracic radiology, respectively) performed the estimations on the institutional digital database system (Vue PACS, version 11.3.5.8902, Carestream Health, Canada) and a consensus was reached after their discussion if there was a disagreement.

Chest CT Evaluation Using Deep Learning-based Quantification

The deep learning-based quantification was performed using a novel established inflammation module (COVID-Lesion Net) based on one automatic segmentation software (Yitu CT, YITU Healthcare Technology Co., Ltd., China). This module was developed as a combination of U-net and Fully convolutional networks ^18,19. In order to detect the lung lesions effectively, a contracting path and an expansive path were employed in this COVID-Lesion Net structure, which consists of three different network components: (1) Twelve convolutional segment, which included convolutional layer, batch normalization layer and an activation layer; (2) Three max-pooling layer for down-sampling; and (3) Three transpose convolutional layer for up-sampling (Fig. 1).. Information on the input CT images was passed through convolutional segments along the two paths. In addition, concatenation operations were performed between convolutional segments as bridges of contracting and expansive paths to improve the information propagation within the network. In order to train the COVID-Lesion Net, chest CT images from 730 patients with confirmed COVID–19 and 1013 patients with normal CT findings in this hospital were retrospectively collected as a training data set (patients not involved in this study). Afterwards, chest CT images from another 212 patients with confirmed COVID–19 and 327 patients with normal CT findings were included as the test data set (patients not involved in this study). 100 training epochs were performed for networking training with a batch size of 8. Adam algorithm was used for the model optimizer. The ground truth region of interest (GT-ROI) for lung lesion was first drawn by a radiologist (LL with 5-year experience in thoracic radiology) and then reviewed by a senior radiologist (GC with 28-year experience in thoracic radiology), who was responsible to modify ROIs if not accepted. Dice coefficient was used to evaluate the performance of this in-house built network for both training and test set using the following equation:

PR-ROI is the predicted ROI drawn by COVID-Lesion Net and the GT-ROI is the ROI drawn by radiologists. As a result, the Dice coefficient is 85.00% for the training set and 82.08% for the test set.

After the lesion detection, Hellinger distance and intersection over union (IOU) of lung CT distribution were calculated to reflect the differences between patients with COVID–19 and reference patients (normal CT findings in the training set) ^20,21. Quantification parameters related to lung lesions including GGO and consolidation were determined by the CT value thresholds ²². The bilateral lungs were also segmented by adaptive thresholding and morphological operation ^23,24. Afterwards, the volumes of bilateral lungs and pulmonary lesions including GGO, consolidation, and both were calculated. In the meanwhile, the percentages of GGO, consolidation, and both (equal to 100 × lesions volume/bilateral lung volume) were calculated as a result of “percent of GGO/consolidation/pulmonary lesions”.

Study Goals

Correlation between conventional CT scoring and the deep learning-based quantification;
Exploring the dynamic patterns using conventional CT scoring and the deep learning-based quantification between moderate and severe groups.

Statistical Analysis

Statistical analyses were performed using IBM SPSS Statistics Software (version 24; IBM, New York, USA). Quantitative data were presented as median with inter-quartile range (IQR) and frequency data were presented as the percentage of the total. The Spearman’s correlation coefficient between CT score and deep learning-based quantification assessed using deep learning-based quantification was calculated. The SPSS curve estimation module was performed to explore the optimal fitting ⁸. A p-value of <0.05 was defined as having statistical significance.

1. Clinical Characteristics

The details of the patients’ clinical information were summarized in Table 1. The median age of the patients was 45y (IQR: 35–60y) with an approximately 1:1 ratio of male to female, but the median age of severe patients was higher than moderate patients (55y vs. 39y). After a median of 8 days (IQR: 5–11 days) from symptom onset, patients were hospitalized. On admission, multiple abnormal biochemical and hematological parameters were observed in severe group, such as decreased lymphocyte count and elevated C-reactive protein. The median hospitalized period was significantly longer in severe group than in moderate group (29 days vs. 18 days). All patients underwent a median of 5 serial chest CT scans (IQR: 4–6) with a median interval of 8 days (IQR: 5–14) in the course.

2. Correlation between conventional CT scoring and deep learning-based quantification

All 465 chest CT scans including 319 chest CT scans in moderate group and 146 chest CT scans in severe group were analyzed using conventional CT scoring and deep learning-based quantification. Based on the analysis of deep learning-based quantification, the involved patients demonstrated significant differences from the reference patients with the median Hellinger distance of 0.24 (IQR: 0.20–0.31) and the median intersection over union (IOU) of 0.66 (IQR: 0.55–0.77). In addition, GGO was identified as the major abnormal finding (median volume of 54.59cm³) (Table 2). The Spearman’s correlation coefficient between CT score and percent of pulmonary lesions assessed by deep learning-based quantification was 0.920 (p<0.001) (Table 3).. Besides, the curve estimation presented an optimal quadratic fitting between two assessments with the r² = 0.924, which was better than the linear fitting (r² = 0.850) (Fig. 2)..

3. Comparisons of conventional CT scoring and deep learning-based quantification between moderate and severe groups at different time points

The severe group presented significantly larger pulmonary lesions indicated as higher CT score and percent of pulmonary involvement calculated using deep learning-based quantification than moderate group at each time point (p<0.001, each) (Table 4).. Besides, the volume of bilaterally uninvolved lungs was significantly lower in severe patients compared to moderate group (Table 4).. In each group at different time points, it demonstrated significant correlations between CT score and percent of pulmonary involvement assessed by deep learning-based quantification (p<0.001) (Table 5).. However, the Spearman’s correlation coefficient was higher in severe group than in moderate group at each time point (Table 5)..

4. Dynamic patterns between moderate and severe groups estimated

CT scoring and the deep learning-based quantification involving 319 chest CT scans in moderate group and 146 chest CT scans in severe group were analyzed using SPSS curve estimations, respectively. Similar patterns were observed for both groups between the predicted CT score and the percentage of pulmonary lesions calculated by deep learning-based quantification (Fig. 3A and 3B).. The pulmonary involvement increased more rapidly and reached the peak at 23^rd days from symptom onset in severe group, while, in moderate group, it reached the peak at 18^th days and experienced faster absorption (Fig. 3A and 3B).. In moderate group, the predicted percentage of GGO and consolidation lesions followed similar patterns, which reached a peak at 18^th days from symptom onset (2.65% and 0.72%, respectively) and decreased afterwards (Fig. 3C and 4).. But in severe group, the peaks of the predicted percentage of GGO and consolidation lesions (23.03% and 4.99%, respectively) were higher than moderate group and the consolidation started to be absorbed earlier than GGO lesions (19 days vs. 23 days from symptom onset) (Fig. 3D)..

This study preliminarily compared a novel deep learning-based qualification to the conventional scoring system in the evaluation of COVID–19 CT manifestations. The results indicated a good correlation between these two estimations and similar findings of the CT patterns between moderate and severe COVID–19, although the correlation was relatively lower in moderate group at different time points than in the severe group. The deep learning-based qualification could calculate the percentage of the lesions separately for GGO and consolidation, which provided an added tool when compared to the conventional scoring system.

In previous studies, the CT demonstrations of COVID–19 evolved through time from symptom onset ^7,8. For example, the GGO was the major early abnormal findings but consolidation was increasingly observed with time till the start of recovery ^7,8. Therefore, irregular chest CT scans of the patients might affect the longitudinal correlation analysis between the conventional CT scoring and the deep learning-based quantification. To avoid this potential impact, only the recovered patients that had experienced serial chest CT scans with relatively regular intervals (median: 8 days) were involved. As a result, 95 patients with serial CT follow-up for more than one month were involved. In consistence to the previous study, severe patients presented elder age and more abnormalities of the laboratory parameters ^25–27. Besides, moderate patients underwent more chest CT scans than severe patients resulting from the statistically longer follow-up period in moderate group compared to the severe group. However, the median interval between two adjacent chest CT scans was the same for both groups. In addition, no significant difference in the period from symptom onset to admission was found between two groups, while the severe patients presented significantly longer hospitalized period owing to the treatment requirements.

All the data of 465 chest CT scans were involved in the correlation analysis between conventional CT scoring and novel deep learning-based quantification. The results demonstrated a good correlation between these two estimations, not only the Spearman’s correlation analysis (r = 0.920, p<0.001). Moreover, the optimal fitting resulted in a quadratic equation (r² = 0.924), which was nearly linear with a relatively low slope when the CT score was less than 5 points. This may imply the risk of over-estimation of lesion areas using conventional CT scoring when the lesions were very small but distributed in multiple lobes. For instance, if there was a very small GGO in each lobe, the CT score might be 5 points, while the deep learning-based quantification could yield a lower value with higher precision. As evidence, it demonstrated a higher correlation between two methods when estimating the severe group, which presented more rapid progression and more extensive pulmonary involvement compared with moderate COVID–19 (peak percent of pulmonary lesions: 27.91% vs. 3.37%) leading to a longer disease course until the radiological resolution.

Another advantage of the deep learning-based quantification was the quantification of the lung volume and the percent of the lung involvements for different types of lesions, which was previously impossible in the context of conventional estimation by radiologists due to the extended workload, especially when mixed lesions were presented ^7,8. The novel quantification modality has enabled the dynamic pattern analysis in different groups with the precise quantification of both GGO and consolidation ²². The quantification results of the dynamic patterns of the moderate and severe patients were similar to a cubic fitting in a previous study ⁸. Furthermore, the results demonstrated that severe patients presented significantly lower lung volume than moderate patients at each time point, which might be attributed to the impairment of pulmonary function caused by COVID–19 or age factor. Therefore, the volume of bilateral lungs might correlate with the COVID–19 severities worth further exploration. On the other hand, although the predicted percent of consolidation reached the peak at the similar time (18–19 days from symptom onset) in both moderate and severe groups, the predicted peak percent of GGO and total pulmonary lesions was delayed in severe group (23 days from symptom onset, each). It was speculated that the absorption of the large area of consolidation might be accompanied by a temporal increase of GGO, reported as the “melting sugar” sign, which simultaneously demonstrated the decrease of solid components and the increase of the lesion area ¹. This phenomenon was not typical in moderate patients where the dynamic changes of GGO and consolidation seemed more synchronized. This was the major difference in the absorption stage between the two severities.

There are limitations in this study. First, although the conventional CT score that was widely used in the CT estimation of COVID–19 was chosen as the reference, to date there has been no gold standard for the lesion area quantification for viral pneumonia. Thus, whether the deep learning-based quantification is more accurate than the CT score is still uncertain. Second, all the deep-learning training and validation were from this single-center, not multi-center. Therefore, more samples from more centers are necessary for further model training to make a better model establishment.

In summary, this study evaluated a novel deep learning-based quantification for COVID–19, which showed a good correlation with the conventional CT scoring. The results indicated the potential application of deep learning-based quantification in the estimation of CT patterns and disease severities for COVID–19, and, in a broader field of view, for other types of viral pneumonia as well.

Acknowledgements

We express our sincere gratitude to the emergency services, nurses, doctors and a lot of medical supports from other provinces for their efforts to combat the COVID–19 outbreak in Wuhan. And we are grateful to YITU Healthcare Technology for providing partial artificial intelligence-related technology support. Besides, we are grateful to Dr. Jiazheng Wang (MSC Clinical & Technical Solutions, Philips Healthcare, Beijing, 100000, China) and Dr. Dandan Zheng (MSC Clinical & Technical Solutions, Philips Healthcare, Beijing, 100000, China) for many useful suggestions through the formation and design of this work.

Author Contributions

Feng Pan and Lin Li contributed equally to this article. Feng Pan, Bo Liang, Lian Yang, and Chuansheng Zheng designed and performed this study. Feng Pan and Lin Li wrote the initial manuscript together with all authors providing critical feedback and edits to subsequent revisions. Bo Liu and Zezhen Ding developed and improved the COVID-Lesion net module. Guangfeng Chen and Tianhe Ye performed the pulmonary lesion segmentation in the training and test sets. Bo Liang, Tianhe Ye, Lingli Li, and Dehan Liu carried out the retrieving work and information input including clinical, radiological, and laboratory data from the institutional electronic database. Lian Yang and Bo Liang take responsibility for the integrity of the data and the accuracy of the data analysis. Bo Liang and Lian Yang were blinded to perform the semi-quantitative analysis of the radiological images on the Vue PACS system. Feng Pan and Lin Li finished the statistical data analysis, and Lian Yang, Bo Liang, and Chuansheng Zheng performed the proofreading work. All authors approved the final draft of the manuscript and agreed to be 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.

Conflicts of Interests

This study was funded by the National Natural Science Foundation of China (NSFC) (No. 81873919) and Rapid Response Key Project of COVID–19 in Huazhong University of Technology (HUST) (No.2020kfyXGYJ021), which was set up for controlling the pandemic of COVID–19 without any financial interests. Dr. Bo Liu and Zezhen Ding are the employees in YITU Healthcare Technology Co., Ltd. The other co-authors declared no conflict of interest, financial or otherwise.

Ethical approval

This retrospective study was approved by the Ethics of Committees of Union Hospital, Tongji Medical College, Huazhong University of Science and Technology (No. 2020–0026), and followed the 1964 Helsinki Declaration and its later amendments or comparable ethical standards.

Patient and other consents

Informed consent/deceased patient permission form for this retrospective study was waived by Ethics of Committees of Union Hospital, Tongji Medical College, Huazhong University of Science and Technology. Only the anonymous data was collected and analyzed to facilitate the radiological diagnosis and grading of COVID–19.

Data Availability Statement

The datasets generated in the current study are available from the corresponding author on request.

1 Pan, Y. et al. Initial CT findings and temporal changes in patients with the novel coronavirus pneumonia (2019-nCoV): a study of 63 patients in Wuhan, China. Eur Radiol 30, 3306-3309, doi:10.1007/s00330-020-06731-x (2020).

2 World Health Organization. Novel coronavirus (2019-nCoV). Situation report 87. Geneva, Switzerland: World Health Organization; 2020. . https://www.who.int/docs/default-source/coronaviruse/situation-reports/20200416-sitrep-87-covid-19.pdf?sfvrsn=9523115a_2 (Published on April 16, 2020).

3 Fang, Y. et al. Sensitivity of Chest CT for COVID-19: Comparison to RT-PCR. Radiology 0, 200432, doi:10.1148/radiol.2020200432 (2020).

4 Xie, X. et al. Chest CT for Typical 2019-nCoV Pneumonia: Relationship to Negative RT-PCR Testing. Radiology 0, 200343, doi:10.1148/radiol.2020200343 (2020).

5 Ai, T. et al. Correlation of Chest CT and RT-PCR Testing in Coronavirus Disease 2019 (COVID-19) in China: A Report of 1014 Cases. Radiology 0, 200642, doi:10.1148/radiol.2020200642 (2020).

6 Wang, Y. et al. Temporal Changes of CT Findings in 90 Patients with COVID-19 Pneumonia: A Longitudinal Study. Radiology, 200843, doi:10.1148/radiol.2020200843 (2020).

7 Shi, H. et al. Radiological findings from 81 patients with COVID-19 pneumonia in Wuhan, China: a descriptive study. Lancet Infect Dis 20, 425-434, doi:10.1016/S1473-3099(20)30086-4 (2020).

8 Pan, F. et al. Time Course of Lung Changes at Chest CT during Recovery from Coronavirus Disease 2019 (COVID-19). Radiology 295, 715-721, doi:10.1148/radiol.2020200370 (2020).

9 Zhao, W., Zhong, Z., Xie, X., Yu, Q. & Liu, J. Relation Between Chest CT Findings and Clinical Conditions of Coronavirus Disease (COVID-19) Pneumonia: A Multicenter Study. AJR Am J Roentgenol 214, 1072-1077, doi:10.2214/AJR.20.22976 (2020).

10 Wu, J. et al. Chest CT Findings in Patients With Coronavirus Disease 2019 and Its Relationship With Clinical Features. Invest Radiol 55, 257-261, doi:10.1097/RLI.0000000000000670 (2020).

11 Li, K. et al. The Clinical and Chest CT Features Associated With Severe and Critical COVID-19 Pneumonia. Invest Radiol 55, 327-331, doi:10.1097/RLI.0000000000000672 (2020).

12 National Health Commission of the People’s Republic of China. Diagnosis and treatment protocols of pneumonia caused by a novel coronavirus (trial version 7). http://www.nhc.gov.cn/yzygj/s7653p/202003/46c9294a7dfe4cef80dc7f5912eb1989/files/ce3e6945832a438eaae415350a8ce964.pdf (Published on March 3, 2020).

13 World Health Organization. Clinical management of severe acute respiratory infection when novel coronavirus (nCoV) infection is suspected. https://apps.who.int/iris/rest/bitstreams/1272156/retrieve (Published on March 13, 2020).

14 Franquet, T. Imaging of pulmonary viral pneumonia. Radiology 260, 18-39, doi:10.1148/radiol.11092149 (2011).

15 Koo, H. J. et al. Radiographic and CT Features of Viral Pneumonia. Radiographics : a review publication of the Radiological Society of North America, Inc 38, 719-739, doi:10.1148/rg.2018170048 (2018).

16 Hansell, D. M. et al. Fleischner Society: glossary of terms for thoracic imaging. Radiology 246, 697-722, doi:10.1148/radiol.2462070712 (2008).

17 Chang, Y. C. et al. Pulmonary sequelae in convalescent patients after severe acute respiratory syndrome: evaluation with thin-section CT. Radiology 236, 1067-1075, doi:10.1148/radiol.2363040958 (2005).

18 Ronneberger, O., Fischer, P. & Brox, T. in medical image computing and computer assisted intervention. 234-241.

19 Long, J., Shelhamer, E. & Darrell, T. in computer vision and pattern recognition. 3431-3440.

20 Rezatofighi, H. et al. in computer vision and pattern recognition. 658-666.

21 Kitsos, C. & Toulias, T. Hellinger Distance Between Generalized Normal Distributions. British Journal of Mathematics & Computer Science 21, 1-16 (2017).

22 Cohen, J. G. et al. Software performance in segmenting ground-glass and solid components of subsolid nodules in pulmonary adenocarcinomas. Eur Radiol 26, 4465-4474, doi:10.1007/s00330-016-4317-3 (2016).

23 Lin-Yu, T. & Li-Chin, H. in AFRICON 2009. 1-5.

24 Pu, J. et al. Adaptive border marching algorithm: automatic lung segmentation on chest CT images. Comput Med Imaging Graph 32, 452-462, doi:10.1016/j.compmedimag.2008.04.005 (2008).

25 Epidemiology Working Group for Ncip Epidemic Response, C. C. f. D. C. & Prevention. [The epidemiological characteristics of an outbreak of 2019 novel coronavirus diseases (COVID-19) in China]. Zhonghua Liu Xing Bing Xue Za Zhi 41, 145-151, doi:10.3760/cma.j.issn.0254-6450.2020.02.003 (2020).

26 Guan, W. J. et al. Clinical Characteristics of Coronavirus Disease 2019 in China. N Engl J Med 382, 1708-1720, doi:10.1056/NEJMoa2002032 (2020).

27 Wang, D. et al. Clinical Characteristics of 138 Hospitalized Patients With 2019 Novel Coronavirus-Infected Pneumonia in Wuhan, China. JAMA, doi:10.1001/jama.2020.1585 (2020).

Table 1. Basic characteristics and clinical outcomes.

	Normal reference	Total, n=95.	Moderate group, n=61.	Severe group, n=34.
Age (y)		45 (35-60)	39 (32-52)	55 (46-68)
Sex
Male		47 (49.5)	27 (44.3)	20 (58.8)
Female		48 (50.5)	34 (55.7)	14 (41.2)
Medical history
Hypertension		10 (10.5)	4 (6.6)	6 (17.6)
Diabetes		5 (5.3)	1 (1.6)	4 (11.8)
Coronary heart disease		1 (1.1)	0 (0.0)	1 (2.9)
Initial symptoms
Fever		81 (85.3)	50 (82.0)	31 (91.2)
Low-grade fever (37.5 - 38.0 ℃)		23 (24.2)	17 (27.9)	6 (17.6)
Moderate fever (38.1 - 39.0 ℃)		35 (36.8)	23 (37.7)	12 (35.3)
High-grade fever (>39.1 ℃)		23 (24.2)	10 (16.4)	13 (38.2)
Cough		58 (61.1)	34 (55.7)	24 (70.6)
Fatigue		51 (53.7)	35 (57.4)	16 (47.1)
Expectoration		23 (24.2)	12 (19.7)	11 (32.4)
Chest distress		13 (13.7)	6 (9.8)	7 (20.6)
Laboratory investigations on admission
White blood cell (G/L)	(3.50-9.50)	4.93 (3.91-6.52)	4.73 (3.78-5.84)	5.77 (4.35-7.22)
Lymphocyte (G/L)	(1.10-3.20)	1.20 (0.84-1.59)	1.36 (1.00-1.79)	0.85 (0.45-1.23)
Hemoglobin (g/L)	(115-150)	129 (121-142)	132 (122-144)	126 (116-136)
Platelet (G/L)	(125-350)	180 (142-242)	177 (150-232)	185 (120-259)
C-reactive protein (mg/L)	(0.00-8.00)	11.90 (4.45-41.92)	7.67 (3.54-15.57)	47.63 (29.18-87.73)
Total bilirubin (μmol/L)	(3.0-20.0)	9.6 (8.3-13.9)	9.5 (8.3-12.2)	10.4 (8.2-15.7)
Alanine aminotransferase (U/L)	(5-35)	34 (22-49)	26 (17-41)	44 (34-66)
Aspartate aminotransferase (U/L)	(8-40)	29 (22-43)	24 (20-31)	44 (33-73)
Lactate dehydrogenase (U/L)	(109-245)	244 (199-377)	201 (161-227)	377 (305-520)
Serum creatinine (μmol/L)	(41.0-81.0)	71.1 (55.5-86.4)	69.5 (55.5-84.5)	71.5 (61.1-88.0)
D-dimer (mg/L)	(<0.5)	0.40 (0.22-1.15)	0.27 (0.22-0.57)	0.77 (0.33-2.15)
Numbers of chest CT scans		5 (4-6)	5 (4-6)	4 (4-5)
Interval between two adjacent chest CT scans (d)		8 (5-14)	8 (5-14)	8 (5-13)
Interval between symptom onset and hospitalization (d)		8 (5-11)	8 (5-11)	10 (5-12)
Hospitalized period (d)		20 (14-30)	18 (11-24)	29 (19-34)
Period of CT follow-up from symptom onset (d)		48 (35-60)	53 (40-66)	38 (32-50)

Table 2. The results of deep learning-based quantification.

	Value (n=465 scans)
Hellinger distance	0.24 (0.20-0.31)
Intersection over union (IOU)	0.66 (0.55-0.77)
Volume of bilateral lungs (cm³)	4036.78 (3377.81-4800.65)
Volume of pulmonary lesions (cm³)	69.85 (2.97-382.31)
Volume of GGO lesions (cm³)	54.59 (2.68-304.00)
Volume of consolidation lesions (cm³)	6.17 (0.25-47.42)
Percent of pulmonary lesions (%)	1.63 (0.06-10.94)
Percent of GGO lesions (%)	1.27 (0.06-8.96)
Percent of consolidation lesions (%)	0.15 (0.01-1.29)

Table 3. Estimations of Spearman’s correlation between CT scoring and deep learning-based quantification.

	Value (n=465 scans)	Spearman’s Correlation coefficient	*p value*
Numbers of involved lobes - Radiologists	3 (1-4)	0.959	<0.001
Numbers of involved lobes - Deep learning-based quantification	3 (1-5)	0.959	<0.001
CT score	3 (1-8)	0.920	<0.001
Percent of pulmonary lesions (%)	1.63 (0.06-10.94)	0.920	<0.001

Note: †Controlling for the variable of "time from symptom onset (d)".

Table 4. Comparisons of CT score and deep learning-based quantification between moderate and severe groups at different time points.

	Moderate group, n=61 patients.	Severe group, n=34 patients.	*p value*
On admission
Conventional CT score	3 (1-5)	11 (5-16)	<0.001
Deep learning-based quantification
Volume of bilateral lungs (cm³)	4002.96 (3459.99-4805.01)	3378.19 (2947.54-4289.97)	0.016
Volume of pulmonary lesions (cm³)	51.35 (5.26-127.96)	664.50 (105.27-1238.84)	<0.001
Volume of GGO lesions (cm³)	44.58 (3.91-97.92)	505.48 (87.23-1133.06)	<0.001
Volume of consolidation lesions (cm³)	7.26 (0.54-28.42)	122.46 (24.44-225.26)	<0.001
Percent of pulmonary lesions (%)	1.38 (0.12-3.18)	19.14 (3.86-40.02)	<0.001
Percent of GGO lesions (%)	1.24 (0.11-2.83)	14.06 (2.82-35.21)	<0.001
Percent of consolidation lesions (%)	0.18 (0.01-0.83)	3.87 (0.47-7.24)	<0.001
1 week after admission
Conventional CT score	3 (2-6)	13 (11-15)	<0.001
Deep learning-based quantification
Volume of bilateral lungs (cm³)	3947.78 (3479.82-4805.71)	3395.71 (2750.45-4043.49)	0.003
Volume of pulmonary lesions (cm³)	89.09 (33.37-213.00)	953.32 (608.28-1283.45)	<0.001
Volume of GGO lesions (cm³)	76.55 (25.35-168.05)	751.15 (529.48-1056.68)	<0.001
Volume of consolidation lesions (cm³)	13.16 (1.70-46.48)	155.00 (72.00-239.90)	<0.001
Percent of pulmonary lesions (%)	2.31 (0.93-5.34)	28.25 (17.35-34.29)	<0.001
Percent of GGO lesions (%)	1.85 (0.78-4.13)	22.25 (15.19-28.53)	<0.001
Percent of consolidation lesions (%)	0.36 (0.04-1.01)	5.17 (2.27-6.87)	<0.001
2 weeks after admission
Conventional CT score	2 (1-3)	11 (7-12)	<0.001
Deep learning-based quantification
Volume of bilateral lungs (cm³)	4220.66 (3638.96-5036.93)	3390.88 (2828.49-4318.19)	<0.001
Volume of pulmonary lesions (cm³)	16.98 (0.26-108.19)	635.73 (324.56-1046.70)	<0.001
Volume of GGO lesions (cm³)	16.47 (0.25-91.10)	549.74 (311.01-863.06)	<0.001
Volume of consolidation lesions (cm³)	1.35 (0.02-6.17)	47.67 (27.23-142.34)	<0.001
Percent of pulmonary lesions (%)	0.46 (0.01-2.25)	18.69 (10.52-28.59)	<0.001
Percent of GGO lesions (%)	0.41 (0.01-1.85)	17.25 (8.74-24.65)	<0.001
Percent of consolidation lesions (%)	0.03 (0.00-0.17)	1.65 (0.72-4.51)	<0.001

Table 5. Correlation at different time points in moderate and severe groups.

	Moderate group, n=61 patients.			Severe group, n=34 patients.
	Value	Spearman’s Coefficient (r)	*p value*	Value	Spearman’s Coefficient (r)	*p value*
On admission
CT score	3 (1-5)	0.749	<0.001	11 (5-16)	0.954	<0.001
Percent of pulmonary lesions (%)	1.38 (0.12-3.18)	0.749	<0.001	19.14 (3.86-40.02)	0.954	<0.001
1 week after admission
CT score	3 (2-6)	0.702	<0.001	13 (11-15)	0.836	<0.001
Percent of pulmonary lesions (%)	2.31 (0.93-5.34)	0.702	<0.001	28.25 (17.35-34.29)	0.836	<0.001
2 weeks after admission
CT score	2 (1-3)	0.788	<0.001	11 (7-12)	0.830	<0.001
Percent of pulmonary lesions (%)	0.46 (0.01-2.25)	0.788	<0.001	18.69 (10.52-28.59)	0.830	<0.001

Download PDF

Version 1

posted

You are reading this latest preprint version

A novel deep learning-based quantification of serial chest computed tomography in Coronavirus Disease 2019 (COVID-19)

Status:

Version 1

Abstract

Figures

Introduction

Materials and Methods

Patients and Groups

CT Scan Protocol

Chest CT Estimation by Radiologists

Chest CT Evaluation Using Deep Learning-based Quantification

Study Goals

Statistical Analysis

Results

1. Clinical Characteristics

3. Comparisons of conventional CT scoring and deep learning-based quantification between moderate and severe groups at different time points

Discussion

Declarations

Acknowledgements

Author Contributions

Conflicts of Interests

Ethical approval

Patient and other consents

Data Availability Statement

References

Tables

Status:

Version 1