Segmentation of Lung Lobes and Lesions in Chest   CT for the Classification of COVID-19 Severity

doi:10.21203/rs.3.rs-2466037/v1

Download PDF

Article

Segmentation of Lung Lobes and Lesions in Chest CT for the Classification of COVID-19 Severity

https://doi.org/10.21203/rs.3.rs-2466037/v1

This work is licensed under a CC BY 4.0 License

Journal Publication

published 28 Nov, 2023

Read the published version in Scientific Reports →

You are reading this latest preprint version

To precisely determine the severity of COVID-19-related pneumonia, computed tomography (CT) is an imaging modality beneficial for patient monitoring and therapy planning. Thus, we aimed to develop a deep learning-based image segmentation model to automatically assess lung lesions related to COVID-19 infection and calculate the total severity score (TSS). The entire dataset consists of 100 COVID-19 patients acquired from Chulabhorn Hospital, divided into 25 cases without lung lesions and 75 cases with lung lesions categorized severity by radiologists regarding TSS. The model combines a 3D-UNet with pre-trained DenseNet and ResNet models for lung lobe segmentation and calculation of the percentage of lung involvement related to COVID-19 infection as well as TSS measured by the Dice similarity coefficient (DSC). Our final model, consisting of 3D-UNet integrated with DenseNet169, achieved segmentation of lung lobes and lesions with Dice similarity coefficients of 0.929 and 0.842, respectively. The calculated TSSs are similar to those evaluated by radiologists, with an R2 of 0.833. The correlation between the ground-truth TSS and model prediction was greater than that of the radiologist, which was 0.993 and 0.836, respectively.

Biological sciences/Computational biology and bioinformatics

Biological sciences/Computational biology and bioinformatics/Image processing

Biological sciences/Computational biology and bioinformatics/Machine learning

Health sciences/Health care/Medical imaging

Health sciences/Health care/Medical imaging/Tomography/Computed tomography

The rapid pandemic-level outbreak of coronavirus disease 2019 (COVID-19) has caused a wide range and degree of illnesses, predominated by respiratory tract infection^1,2,3,4. Although most infected patients show asymptomatic or mild clinical manifestations, further investigation beyond real-time reverse transcriptase polymerase chain reaction (RT-PCR) or rapid Covid-19 tests such as chest radiographs is routinely indicated in worsening cases that require hospitalization^5,6. Characteristic findings in chest radiographs of COVID-19 related pneumonia are bilateral patchy and/or confluent and bandlike ground-glass opacity or consolidation in a peripheral and mid-to-lower lung zone distribution. By contrast, several studies have found almost one-half of normal chest radiographs at initial presentation disagree with clinical symptoms^7,8,9,10.

Because of its higher sensitivity, specificity, and speed, chest computed tomography (CT) has become more useful than RT-PCR in early detection, to obtain more information about chest pathology, and to evaluate the severity of lung involvement. Moreover, it can assist triage, especially when hospitalization is required but there is a shortage of healthcare personnel, inpatient beds, and medical equipment, and it may be useful as a standard modality for the rapid diagnosis of Covid-19 related pneumonia^{11,12,13,14,15}. The chest CT findings are peripheral, bilateral, ground-glass opacity(GGO) with some round shapes with or without consolidation or intralobular lines; or reverse halo sign or other findings of organizing pneumonia^16,17,18,19. The total severity score (TSS) has been proposed by Chung et al. ²⁰ which is calculated from summation of lesions in five lung lobes and is used to categorize the severity of lung involvement and help determine the proper therapeutic management and prognosis²¹.

To reduce in the amount of time required for interpretation and increase in the accuracy with which lesions can be detected, deep learning has been used to efficiently analyze medical images by performing tasks such as semantic segmentation, which is image annotation at the pixel level. The outcome of classifying each pixel is a target area in an image. The U-Net model is a convolutional neural network-based model that was originally used for the semantic segmentation of biomedical images and is now one of the most utilized image segmentation techniques. The model structure is U-shaped and consists of two parts: a contracting path (encoder) and an expanding path (decoder)²². Subsequently, a U-Net model was created to support three-dimensional (3D) matrices and is called 3D-Unet²³. The 3D-UNet model was used to develop a more efficient 3D imaging model for the segmentation of lesions and lung tissue^24,25. Cropping the lung area before lesion segmentation can improve accuracy²⁶. Enshaei et al.²⁷ developed a model for predicting the lesion area of COVID-19 patients from CT-scan images, using a model to predict the lung area before the lesion regions were considered. This method enables the lesion model to predict lesions more accurately. In another study, a deep learning model was applied to lung lobe segmentation. The model is capable of accurately segmenting each lung lobe from lung CT scans²⁸. It is also utilized in lung lobe segmentation analysis for lung segmentation research to improve segmentation accuracy in multiple diseases such as chronic obstructive pulmonary disease (COPD), lung cancer, and COVID-19 related pneumonia²⁹.

As an example of the use of a deep learning model for computer-aided diagnostics to determine the intensity of an infection, Aswathy A. L. and Vinod Chandra S. S.³⁰ used 3D-UNet models to segment the lung parenchyma and infected regions in lung CT scans. The models obtained accurate segmentations. Qiblawey et al.³¹ developed a program to classify the severity of lung CT in COVID-19 patients, using deep learning techniques to analyze the lesion area and compare it with the lung area in lung CT scan images. They calculated the percentage of infection (PI) using a U-Net model combined with pre-trained models such as ResNet and DenseNet. The residual neural network (ResNet) was first presented by He et al.³² as a solution to the vanishing gradient problem. To do this, they added a residual block (that records X values across layers for co-computation in the next layer) findings of organizing pneumonia. DenseNet was first presented by Huang et al.³³ The model structure consists of dense blocks, each containing a convolutional layer. This structure enables the model to learn the features of the data. For this reason, this knowledge can be applied to lung lobe segmentation and lesion segmentation in CT scan images.

In this study, deep learning semantic segmentation was used for the lung severity scoring of COVID-19 infection. The proposed method utilizes a combination of 3D-UNet models integrated with pre-trained models, DenseNet and ResNet, to compute the PI from the lung lobe and lesion segmentation results and estimate the TSS automatically. The aim is to alleviate the radiologist's workload and time spent on imaging diagnostics as well as improve reporting accuracy.

Datasets

Due to retrospective nature, informed consent was waived and all data were anonymized. This project was approved by the human research ethics committee of the Chulabhorn Research Institute (research project code 167/2564) and complied with the declaration of Helsinki. The entire CT dataset consists of 100 COVID-19 patients confirmed by RT-PCR acquired from Chulabhorn Hospital who underwent non-contrast enhanced chest CT as a part of routine clinical care throughout the pandemic. For each patient, an axial chest CT scan dataset was utilized with Hounsfield unit values ranging from − 1500 to -500 and the total number of CT slices ranging from 92 to 208. The data set contained 25 cases without lung lesions and 75 cases with lung lesions categorized as mild, moderate, and severe by experienced radiologists according to TSS and was divided into three sections, as described in Table 1. Model training and validation were performed with the training set. The model segmentation performance was evaluated utilizing test set 1. The TSS evaluation was performed using test set 2. Test sets 1 and 2 include two identical severe cases because Thailand's population has already received vaccines for the current situation.

Table 1

Axial lung CT scan datasets
Case type	Training set			Test set 1			Test set 2
Case type	Number of cases	Total slices	Avg ± std per case	Number of cases	Total slices	Avg ± std per case	Number of cases	Total slices	Avg ± std per case
No lesions	8	985	123 ± 22	2	365	183 ± 19	15	1,953	130 ± 27
Mild	8	907	113 ± 12	2	388	194 ± 4	15	2,008	134 ± 27
Moderate	8	941	118 ± 13	2	260	130 ± 27	26	2,995	115 ± 20
Severe	8	919	115 ± 23	2	301	151 ± 43	6	730	122 ± 32
Total	32	3752	117 ± 18	8	1314	164 ± 34	62	7686	124 ± 25

Data preprocessing

The lung CT data were saved in JPEG format with a resolution of 512 × 512 pixels. The CT images were labeled by a program called LabelME³⁴, and the output labels were in JavaScript Object Notation (JSON) format. All labeled data were validated by four radiologists. The JSON files were converted into matrices for model training.

In the data preprocessing phase, CT scan images (JPEG format) and labeled data (JSON format) were resized using the “cv2.resize” function from 512 × 512 pixels to 256 × 256 pixels to minimize the required memory resources (RAM). The parameter interpolation was set to “INTER_AREA” for the CT scan images and “INTER_NEAREST" for the label data because this solution prevented the values specified in each pixel from being changed. In addition, our model input matrix was fixed to a size of 128 × 256 × 256. The CT volumes were adjusted to 128 images per patient according to the following three rules (Fig. 1).

If the CT volume comprised 128 or fewer images, a 256 × 256 zero-padding matrix was added to increase the volume to 128 images.
If the CT volume had between 129 and 175 CT images, 128 images from the CT volume's middle range were selected to train the model.
If the CT volume contained more than 175 CT images, only odd-numbered images were selected. This strategy halves the number of images. A zero-padding matrix with a resolution of 256 × 256 was added to increase the volume to 128 images.

A color adjustment method was applied to improve image contrast by using the contrast-limited adaptive histogram equalization (CLAHE) technique, which is an OpenCV library³⁵. The CLAHE parameters were set to a clipLimit of 3 and a tileGridSize of (8, 8). Models were trained/tested in two experiments: the first with original images (no color adjustment) and the second with CLAHE-adjusted images.

Model training

Two models were used in this study: (1) a pulmonary lobe segmentation model and (2) a lesion segmentation model. According to related studies, a model that combines a 3D-UNet structure with a DenseNet or ResNet is effective in segmenting parts of the image precisely. Therefore, pre-trained DenseNet121, DenseNet169, DenseNet201, ResNet18, ResNet34, ResNet50, ResNet101, and ResNet152 models were obtained through an application programming interface (API) from Solovyev et al.³⁶

Pulmonary lobe segmentation. A multiclass semantic segmentation model was used to segment the five lung lobes. Annotated labels consisted of six categories 0, 1, 2, 3, 4, and 5, which indicate the background, right upper lobe (RUL), right lower lobe (RLL), right middle lobe (RML), left upper lobe (LUL), and left lower lobe (LLL), respectively.

Lesion segmentation. The lesion model was developed from a binary semantic segmentation model that outputs the value 1 for lesion areas and 0 for background areas. Images without extrapulmonary regions are preferred for lesion model training. The dataset used for model training was preprocessed as described in the data preparation section.

In the model training process, lung lobe and lesions segmentation models were trained on servers equipped with an Intel(R) Xeon(R) Gold 6126 CPU 2.60GHz, 40 GB RAM, and an NVIDIA Tesla V100 SXM2 GPU. Figure 2 shows the overall workflow. The model’s output is the predicted class for each pixel, which is then used to compute the percentage area of lesions in each lung lobe for the CT score. This score is then used to calculate the TSS value for diagnosing the severity of the current pathology. For both models, Adam optimization was used, the loss function was a hybrid loss function (focal loss + Dice loss), the learning rate was set to 0.0001, the L2 regularizer was used with a value of 0.01, the batch size was set to 1, and the maximum number of epochs was 200. The lesion model activation function was set to sigmoid with a dropout rate of 0.4, whereas the pulmonary lobe model activation function was set to softmax with dropout rate of 0.2. The hybrid loss technique³⁷, which combines focal loss and Dice loss, was used to improve model performance.

PI

The PI in each lung lobe is calculated by dividing the number of predicted lesion pixels by the total number of lung lobe pixels in the CT volume. The predicted lesion pixels were obtained from the output of the lesion segmentation model, whereas the predicted lung lobe pixels were derived from the output of the lung lobe segmentation model, in which the value of each pixel identifies the lobar type in the lung CT image. Therefore, the PI is calculated by performing the following equation.

$$Percentage of{ Infection}_{lobe}= \frac{Lesion Area \left(pixels\right)}{Lung Lobe Area \left(pixels\right)}✕100$$

TSS

The TSS proposed by Chung et al. ²⁰ was calculated from the sum of the five lobe CT score, which was calculated from PI based on the criteria listed in Table 2. The severity of COVID-19 patients can be classified from the TSS value based on the severity criteria in Table 3 and the following equation:

$$TSS = {CT Scor{e}_{RUL}+CT Scor{e}_{RML}+CT Scor{e}_{RLL}+CT Scor{e}_{LUL}+CT Scor{e}_{LLL}}_{}$$

Table 2

CT-score criteria based on the percentage of infection (PI) for the diagnosis of pneumonia in COVID-19 patients proposed by Chung et al.²⁰
PI	Score
No lesion	0
< 5%	1
6–25%	2
26–50%	3
51–75%	4
> 75%	5

Table 3

Severity type classification based on the total severity score (TSS)
TSS	Severity type
≤ 7	Mild
8–17	Moderate
≥ 18	Severe

Evaluation metric

The most commonly utilized measurement to evaluate image segmentation is the Dice similarity coefficient (DSC). The DSC calculates the relative overlap between the predicted area and ground truth, and it is used to choose the most appropriate model. The DSC is defined as follows:

$$DSC = \frac{2TP}{2TP + FP + FN}$$

where the term “true positive” (TP) refers to an outcome such that the model correctly predicts the positive class, “false positive” (FP) is an outcome such that the model incorrectly predicts the positive class, and “false negative” (FN) is an outcome such that the model incorrectly predicts the negative class.

Pulmonary lobe segmentation

The testing of lung lobe segmentation models was divided into two sections, one for the models trained on the original images and one for the models trained on the lung CT images processed by the CLAHE technique. The model's test results for test set 1 are given in Supplementary Tables S1 and S2. The 3D-UNet + DenseNet169 model trained with the original CT data was found to obtain the best lung lobe segmentation with a DSC of 0.929 ± 0.013.

For lung lobe segmentation, the results indicate that the model trained on original images outperforms the model trained on the CLAHE-adjusted images. Figure 3 presents the middle range of lung lobe CT scan segmentation results for a model trained on the original images and CLAHE-adjusted images. The lung lobe segmentation results in the no lesion group obtained by the 3D-UNet + DenseNet169 model are better than those of the model trained using CLAHE-adjusted images, as demonstrated by the right middle lobe lung (blue) in Fig. 3. The model erroneously labels the right upper lung (red) as part of the background area. By contrast, the model trained with the original images was more accurate when segmenting the right middle lobe (blue). Table 4 presents the image segmentation results of the lung lobe segmentation evaluation of the 3D-Unet + DenseNet169 model trained on the original images. The segmentation results for the mild group were worse than those for the other groups.

Table 4

Lung lobe segmentation results of the 3D-UNet + DenseNet169 model with test set 1 in each group.
Case type	DSC
Case type	RUL	RML	RLL	LUL	LLL	Avg ± std
No lesion	0.865	0.918	0.945	0.908	0.965	0.920 ± 0.038
Mild	0.905	0.884	0.896	0.932	0.928	0.909 ± 0.021
Moderate	0.930	0.951	0.953	0.932	0.936	0.940 ± 0.011
Severe	0.937	0.950	0.932	0.962	0.949	0.946 ± 0.012
Overall	0.909	0.926	0.931	0.934	0.944	0.929 ± 0.013

Lesion segmentation

The results of the lesion model on test set 1 are divided into two sections, one for the original images and one for the images processed using the CLAHE technique (Supplementary Tables S3 and S4). The models trained with the CT images processed by CLAHE were found to be the most effective for lesion segmentation; the 3D-UNet + DenseNet169 model obtained a DSC of 0.842.

According to the test results, the model trained on the CLAHE-adjusted images performs better than the model trained on the original images. The best model (3D-UNet + DenseNet169) was evaluated on test set 1, and the results are shown in Table 5 and Fig. 4. These results imply that lesion models can often accurately predict lesions when images are completely free of lesions (no-lesion images) or when lesions appear throughout the lungs (severe images).

Table 5

Lesion segmentation results of the 3D-UNet + DenseNet169 model on test set 1 for each severity level.
Severity level	DSC
No lesions	1.000
Mild	0.766
Moderate	0.750
Severe	0.853
Overall	0.842

TSS regression analysis

The best performing lung and lesion models with the highest DSC were applied to perform image segmentation for 62 cases (test set 2). The segmentation results were used to calculate the PI in each lung lobe, which was then used to compute the TSS. To analyze the trends and correlation of the TSS results, we compared the radiologist measurements with TSS values calculated based on our approach using correlation coefficients and R² values. Figure 5 presents the results of a regression analysis of the TSS values obtained by models and those obtained by radiologists. According to the appropriate statistical analysis, the correlation coefficient of 0.913 indicates that the model-based predictions and radiologist's diagnosis are positively correlated with an R² of 0.833.

Additionally, three techniques for measuring the TSS value were compared: using the label mask of test set 1 as the ground truth (GT), the radiologist's diagnostic TSS (radiologist), and the model-based TSS (prediction). TSS (radiologist) indicates the radiologist-determined TSS value. TSS (GT) indicates the TSS value obtained by using the ground truth for the calculation. TSS (prediction) is the TSS result obtained using the model predictions for the computation. The outcomes of the investigation are presented in Fig. 6. The TSS (prediction) values were correlated with TSS (GT) values with an R² of 0.993. The R² of the TSS (GT) and TSS (radiologist) comparison (0.846) indicates that the model is more accurate than a human.

We compared the proposed lung lobe segmentation model's performance with the performance of models in other related studies. Tang et al.³⁷ reported a DSC of 0.915 on the LUNA16 dataset and 0.942 on the Tianchi dataset, indicating that our model underperformed theirs. However, Tang et al. employed CT datasets of lungs without lesions for their test datasets, and the models were trained on 40 cases, whereas our model utilized the no-lesion training dataset, which consists of eight cases. The performance of our model (3D-UNet + DenseNet169) is equivalent to that of Tang et al. with a DSC of 0.920 in the no-lesions group (Table 4). This demonstrates that our approach can be trained using a small dataset. Furthermore, a comparison with a lung segmentation algorithm based on the lung fissure surface yielded a DSC of 0.840³⁸. This demonstrates that a deep learning approach can efficiently segment the lung even when the fissure is not apparent.

For the lesion segmentation studies, Xiao et al.²⁴ reported a 3D-UNet model with a DSC of 0.891; Qiblawey et al.³¹ reported a DSC of 0.941 for Feature Pyramid Network (FPN) combined with DenseNet201 model. Both of these models outperformed our model because training a model requires a considerable variety of data. For example, Qiblawey et al. used a considerable number of images for model training (15,698 images), whereas our model training used a limited number of images (3,752 images). However, our lesion model's results demonstrated that it was capable of accurately predicting small and large lesions in the no-lesion and severe groups, despite the inaccuracies in lesion segmentation for the mild and moderate groups. According to the findings of the experiment with color adjustment methods (CLAHE), the lung model accurately segmented the original images better than it did the CLAHE images. By contrast, the lesion model can segment the CLAHE images better than it can the original images. The CLAHE method emphasizes the white lesion regions, rendering them more apparent, whereas lung fissures are more difficult to detect.

Additionally, the TSS regression analysis showed that our model can be utilized to segment lung CT scans effectively. An analysis of the radiologist's diagnostic TSS and model based TSS results revealed that our method produced observation results that were more accurate with respect to the ground truth than human observation. This demonstrates that the measurement software could be used to eliminate human error in the estimation of infection rates in the lung area. Furthermore, our approach can perform lesion area segmentation and calculate the infection rate automatically. This could dramatically reduce the radiologist workload and enhance the efficiency of diagnosing COVID-19 severity levels.

Constructing a model for the automatic segmentation and scoring of COVID-19 infection in chest CT was accomplished through the application of deep learning techniques. According to the findings, the combination of 3D-UNet and DenseNet169 achieved the highest level of performance when it came to the classification of lung lobes and lesions. The projected severity score had a strong correlation with the visual assessments made by radiologists. This accurate model provided a dependable method for quantifying the extent of lung involvement. The proposed model is helpful in determining the extent of the lower respiratory tract infection and monitoring of the disease in COVID-19 patients.

Acknowledgements

The authors would like to express our deepest appreciation to everyone who offered advice on this project. We are thankful to the Faculty of Health Science Technology, Princess Chulabhorn College of Medical Science, Chulabhorn Royal Academy, for providing the dataset. We are grateful to the Princess Srisavangavadhana College of Medicine, Chulabhorn Royal Academy for providing knowledge and equipment for this project. A special thanks to the Department of Computer Engineering, Faculty of Engineering, King Mongkut’s University of Technology Thonburi for providing the server used to train the model. Lastly, we thank Kimberly Moravec, PhD, from Edanz (https://www.edanz.com/ac) for editing a draft of this manuscript.

Author contributions

P.K., P.P., T.B., U.C.T., K.C., and N.T. designed the study. T.B. mainly performed data annotation. A.C., N.W., T.J., and N.T. checked and validated the annotation results. P.K. and P.P. developed and analyzed the lung lobe and lesion model. P.K. and P.P. wrote the main manuscript text. U.C.T.; K.C., U.D.T., and N.T. revised the manuscript. All authors reviewed and approved the submitted manuscript.

Competing interests

The authors declare no competing interests.

Data Availability

The datasets used and/or analyzed during the current study available from the corresponding author on reasonable request.

Code availability

The source code used for this research work is made publicly available in the GitHub repository https://github.com/hds-69/csc-app.

Guan, W.-J. et al. Clinical characteristics of Coronavirus disease 2019 in China. N. Engl. J. Med. 382, 1708–1720 (2020).
Z hu, N. et al. A novel Coronavirus from patients with pneumonia in China, 2019. N. Engl. J. Med. 382, 727–733 (2020).
Huang, C. et al. Clinical features of patients infected with 2019 novel coronavirus in Wuhan, China. Lancet 395, 497–506 (2020).
Berlin, D. A., Gulick, R. M. & Martinez, F. J. Severe covid-19. N. Engl. J. Med. 383, 2451–2460 (2020).
ACR recommendations for the use of chest radiography and computed tomography (CT) for suspected COVID-19 infection. Acr.org https://www.acr.org/Advocacy-and-Economics/ACR-Position-Statements/Recommendations-for-Chest-Radiography-and-CT-for-Suspected-COVID19-Infection (2020).
Rubin, G. D. et al. The role of chest imaging in patient management during the COVID-19 pandemic: A multinational consensus statement from the fleischner society. Chest 158, 106–116 (2020).
Wong, H. Y. F. et al. Frequency and distribution of chest radiographic findings in patients positive for COVID-19. Radiology 296, E72–E78 (2020).
Smith, D. L., Grenier, J.-P., Batte, C. & Spieler, B. A characteristic chest radiographic pattern in the setting of the COVID-19 pandemic. Radiol Cardiothorac Imaging 2, e200280 (2020).
Litmanovich, D. E., Chung, M., Kirkbride, R. R., Kicska, G. & Kanne, J. P. Review of chest radiograph findings of COVID-19 pneumonia and suggested reporting language. J. Thorac. Imaging 35, 354–360 (2020).
Weinstock, M. B. Chest x-ray findings in 636 ambulatory patients with COVID-19 presenting to an urgent care center: a normal chest x-ray is no guarantee. J. Urgent Care. Med 14, 13–18 (2020).
Li, Y. & Xia, L. Coronavirus disease 2019 (COVID-19): Role of chest CT in diagnosis and management. AJR Am. J. Roentgenol. 214, 1280–1286 (2020).
Zhou, H. et al. Coronavirus disease 2019 (COVID-19): chest CT characteristics benefit to early disease recognition and patient classification-a single center experience. Ann. Transl. Med. 8, 679 (2020).
Borakati, A., Perera, A., Johnson, J. & Sood, T. Diagnostic accuracy of X-ray versus CT in COVID-19: a propensity-matched database study. BMJ Open 10, e042946 (2020).
Sverzellati, N. et al. Chest radiography or computed tomography for COVID-19 pneumonia? Comparative study in a simulated triage setting. Eur. Respir. J. 58, 2004188 (2021).
Kim, H., Hong, H. & Yoon, S. H. Diagnostic performance of CT and reverse transcriptase polymerase chain reaction for Coronavirus disease 2019: A meta-analysis. Radiology 296, E145–E155 (2020).
Zhou, S., Wang, Y., Zhu, T. & Xia, L. CT features of Coronavirus disease 2019 (COVID-19) pneumonia in 62 patients in Wuhan, China. AJR Am. J. Roentgenol. 214, 1287–1294 (2020).
Bao, C., Liu, X., Zhang, H., Li, Y. & Liu, J. Coronavirus disease 2019 (COVID-19) CT findings: A systematic review and meta-analysis. J. Am. Coll. Radiol. 17, 701–709 (2020).
Niu, R. et al. Chest CT features associated with the clinical characteristics of patients with COVID-19 pneumonia. Ann. Med. 53, 169–180 (2021).
Simpson, S. et al. Radiological society of north America expert consensus statement on reporting chest CT findings related to COVID-19. Endorsed by the society of thoracic radiology, the American college of radiology, and RSNA - secondary publication. J. Thorac. Imaging 35, 219–227 (2020).
Chung, Michael et al. CT Imaging Features of 2019 Novel Coronavirus (2019-nCoV). Radiology vol. 295,1 (2020): 202-207.
Li, K. et al. CT image visual quantitative evaluation and clinical classification of coronavirus disease (COVID-19). Eur. Radiol. 30, 4407–4416 (2020).
Ronneberger, O., Fischer, P. & Brox, T. U-Net: Convolutional Networks for Biomedical Image Segmentation. in Lecture Notes in Computer Science 234–241 (Springer International Publishing, 2015).
Çiçek, Ö., Abdulkadir, A., Lienkamp, S. S., Brox, T. & Ronneberger, O. 3D U-net: Learning dense volumetric segmentation from sparse annotation. in Medical Image Computing and Computer-Assisted Intervention – MICCAI 2016 424–432 (Springer International Publishing, 2016).
Xiao, Z., Liu, B., Geng, L., Zhang, F. & Liu, Y. Segmentation of lung nodules using improved 3D-UNet neural network. Symmetry (Basel) 12, 1787 (2020).
Zhou, L. et al. An interpretable deep learning workflow for discovering subvisual abnormalities in CT scans of COVID-19 inpatients and survivors. Nat Mach Intell 4, 494–503 (2022).
Mansoor, A. et al. A generic approach to pathological lung segmentation. IEEE Trans. Med. Imaging 33, 2293–2310 (2014).
Enshaei, N. et al. COVID-rate: an automated framework for segmentation of COVID-19 lesions from chest CT images. Sci. Rep. 12, 3212 (2022).
Gerard, S. E. & Reinhardt, J. M. Pulmonary lobe segmentation using A sequence of convolutional neural networks for marginal learning. in 2019 IEEE 16th International Symposium on Biomedical Imaging (ISBI 2019). 1207–1211 (IEEE, 2019).
Gerard, S. E. et al. CT image segmentation for inflamed and fibrotic lungs using a multi-resolution convolutional neural network. Sci. Rep. 11, 1455 (2021).
Aswathy, A. L. & Vinod Chandra SS. Cascaded 3D UNet architecture for segmenting the COVID-19 infection from lung CT volume. Sci. Rep. 12, 3090 (2022).
Qiblawey, Y. et al. Detection and severity classification of COVID-19 in CT images using deep learning. Diagnostics (Basel) 11, 893 (2021).
He, K., Zhang, X., Ren, S. & Sun, J. Deep residual learning for image recognition. in 2016 IEEE Conference on Computer Vision and Pattern Recognition (CVPR). 770-778 (IEEE, 2016).
Huang, G., Liu, Z., Van Der Maaten, L. & Weinberger, K. Q. Densely connected convolutional networks. in 2017 IEEE Conference on Computer Vision and Pattern Recognition (CVPR). 2261-2269 (IEEE, 2017).
Russell, B. C., Torralba, A., Murphy, K. P. & Freeman, W. T. LabelMe: A database and web-based tool for image annotation. Int. J. Comput. Vis. 77, 157–173 (2008).
OpenCV: Histograms - 2: Histogram equalization. Opencv.org https://docs.opencv.org/4.x/d5/daf/tutorial_py_histogram_equalization.html (2022).
Solovyev, R., Kalinin, A. A. & Gabruseva, T. 3D convolutional neural networks for stalled brain capillary detection. Comput. Biol. Med. 141, 105089 (2022).
Tang, H., Zhang, C. & Xie, X. Automatic pulmonary lobe segmentation using deep learning. in 2019 IEEE 16th International Symposium on Biomedical Imaging (ISBI 2019). 1225-1228 (IEEE, 2019).
Chen, X., Zhao, H. & Zhou, P. Lung lobe segmentation based on lung fissure surface classification using a point cloud region growing approach. Algorithms 13, 263 (2020).

No competing interests reported.

Supplementary.zip

Download PDF

Journal Publication

published 28 Nov, 2023

Read the published version in Scientific Reports →

Editorial decision: Major revision
17 Mar, 2023
Reviews received at journal
02 Mar, 2023
Reviewers agreed at journal
01 Mar, 2023
Reviewers agreed at journal
01 Mar, 2023
Reviewers invited by journal
01 Mar, 2023
Editor assigned by journal
01 Mar, 2023
Editor invited by journal
17 Jan, 2023
Submission checks completed at journal
17 Jan, 2023
First submitted to journal
11 Jan, 2023

You are reading this latest preprint version

Segmentation of Lung Lobes and Lesions in Chest CT for the Classification of COVID-19 Severity

Status:

Journal Publication

Version 1

Abstract

Figures

Introduction

Materials And Methods

Datasets

Data preprocessing

Model training

PI

TSS

Results

Evaluation metric

Pulmonary lobe segmentation

Lesion segmentation

TSS regression analysis

Discussion

Conclusion

Declarations

References

Additional Declarations

Supplementary Files

Status:

Journal Publication

Version 1