DBCU-Net: Deep Learning Approach for Segmentation of Coronary Angiography Images

DBCU-Net: Deep Learning Approach for Segmentation of Coronary Angiography Images | Research Square window.SnipcartSettings = { analytics: { enabled: false } }; (function() { var accessVector = localStorage.getItem('access_vector') || ''; window.dataLayer = window.dataLayer || []; if (accessVector) { window.dataLayer.push({ user: { profile: { profileInfo: { snid: accessVector } } } }); } })(); (function(w,d,s,l,i){w[l]=w[l]||[];w[l].push({'gtm.start':new Date().getTime(),event:'gtm.js'});var f=d.getElementsByTagName(s)[0],j=d.createElement(s),dl=l!='dataLayer'?'&l='+l:'';j.async=true;j.src='https://www.googletagmanager.com/gtm.js?id='+i+dl;f.parentNode.insertBefore(j,f);})(window,document,'script','dataLayer','GTM-K279D39R'); Browse Preprints In Review Journals COVID-19 Preprints AJE Video Bytes Research Tools Research Promotion AJE Professional Editing AJE Rubriq About Preprint Platform In Review Editorial Policies Our Team Advisory Board Help Center Sign In Submit a Preprint Cite Share Download PDF Research Article DBCU-Net: Deep Learning Approach for Segmentation of Coronary Angiography Images Yuqiang Shen, Zhe Chen, Jijun Tong, Nan Jiang, Yun Ning This is a preprint; it has not been peer reviewed by a journal. https://doi.org/ 10.21203/rs.3.rs-2028133/v1 This work is licensed under a CC BY 4.0 License Status: Published Journal Publication published 05 Apr, 2023 Read the published version in The International Journal of Cardiovascular Imaging → Version 1 posted 11 You are reading this latest preprint version Abstract Coronary angiography (CAG) is the “gold standard” for diagnosing coronary artery disease (CAD). However, due to the limitation of current imaging methods, the CAG image has low resolution and poor contrast with a lot of artifacts and noise, which makes it difficult for blood vessels segmentation. In this paper, we propose a DBCU-Net for automatic segmentation of CAG images, which is an extension of U-Net, DenseNet with bidirectional convLSTM. The main contribution of our network is that instead of convolution in the feature extraction of U-Net, we incorporate dense connectivity and the bidirectional convLSTM to highlight salient features. We conduct our experiment on our private dataset, and achieve average Accuracy, Precision, Recall and F1-score for coronary artery segmentation of 0.985, 0.913, 0.847 and 0.879 respectively. Coronary angiography U-Net Vessel segmentation Deep Learning Figures Figure 1 Figure 2 Figure 3 Figure 4 Figure 5 Figure 6 Figure 7 1 Introduction Coronary artery disease (CAD) is a common heart disease, which is the leading cause of death worldwide according to the World Health Organization (WHO) [ 1 ]. Globally, the number of patients with CAD is expected to increase from 327.9 million in 2017 to 365.9 million in 2026. CAD poses a huge threat to human life. An accurate diagnosis of CAD is particularly important. With the development of computer diagnosis and treatment technologies, more powerful medical methods, such as magnetic resonance imaging(MRI), computed tomography(CT) [ 2 ], and X-ray coronary angiography(CAG)[ 3 ], are proposed to assist diagnosis. As the "gold standard" for the diagnosis of CAD, CAG can precisely pinpoint the site and the degree of coronary artery stenosis, as well as the symptoms of the condition. Accurate medical image analysis is crucial to subsequent clinical diagnosis and treatment. Diagnosis and treatment of clinical diseases mainly rely on advanced instruments and doctors' high-precision technology. However, manual segmentation of such CAG images requires a lot of medical expertise, which is time-consuming and is prone to human error. Due to the shortage of medical resources, a computer diagnosis system that can assist doctors in making better diagnoses and determining follow-up treatment plans in a shorter time is preferred. In normal CAG image segmentation, the images are divided into two parts: vessels and backgrounds. Although traditional machine learning techniques (e.g. model-based methods and atlas-based methods [ 4 , 5 , 6 , 7 , 8 ]) have achieved good performance in vessels segmentation, deep learning methods have defeated traditional techniques with its automation and versatility in terms of segmentation efficiency and detection accuracy. In this paper, we propose a DBCU-Net for automatic segmentation of CAG images, which is an extension of U-Net , D enseNet [ 9 ] with b idirectional c onvLSTM [ 10 , 11 ]. The dense connectivity strengthens feature propagation and gets multi-level feature to enhance representation. The bidirectional convLSTM ensures both forward and backward passes are used simultaneously. The two features of the encoding layer and the decoding layer are combined in each direction to realize the segmentation of blood vessels. The contributions of our works are listed as follows: (1) Many other tissues and noises in the background can be accurately segmented. (2) The problem of missegmentation of vessels in other branches of the heart in the background is improved. (3) The segmentation of tiny vessels in the distal terminals has the same positive outcomes. We conduct our experiment on our private dataset, and obtain competitive performance. 2 Related Work Since 2010, deep learning has outperformed the conventional state-of-the-art approaches towards visual recognition tasks like heart MRI segmentation and cancer lung nodule segmentation. Ciresan et al [ 12 ]. used CNN to automatically segment electron microscope images for the first time, and won the electron microscope challenge with a huge advantage. In 2015, Ronneberger et al [ 13 ]. proposed a groundbreaking network architecture named U-Net. U-Net is composed of an encoder-decoder structure, where the encoding path includes four down-sampling layers, a total of 16 times down-sampling. Symmetrically, its decoding path is up-sampled four times accordingly. The high-level semantic information feature map obtained from the encoding path is restored to the resolution of the original image, and the feature map stitching is performed on the same stage using skipping connection. Finally, a 1x1 convolution is used to obtain the required number of classes (e.g. vessels and background) from each component feature. Although U-Net has achieved good results in medical image segmentation, it also has some limitations. First of all, a large and deep network requires about 19 million parameters. To better match the network, the quantity of date needed for training is large. Secondly, to make the network achieve higher accuracy, more convolutions need to be added to the network, which causes the network has to learn more unnecessary information that leads to overfitting. Finally, as the network deepens, there may be a risk of gradient disappearance, which brings great risks to the training of deep neural networks. To solve the above-mentioned architectural problems, many U-Net-based network deformation structures have been proposed and achieved good results. A new U-Net-based DNN structure was proposed to improve the segmentation effect by Ahmed et al . [ 14 ] This method selected various enhancement methods (e.g. histogram and Frangi multi-scale filtering) according to the background to remove noise and improve segment effect. Shi et al . [ 15 ] proposed an adaptive generative confrontation network to complete the segmentation of coronary blood vessels. This method uses an adaptive U network as a generator and a three-layer pyramid structure as a discriminator. Under the mechanism of generating confrontation, both the generators can extract the fine features of coronary arteries. Compared with other blood vessel segmentation methods, the segmentation accuracy and continuity of this method have been greatly improved. However, for U-Net, this method requires multiple prior information. With the continuous development of deep learning in medical image segmentation, many scholars have combined U-Net with other deep neural networks and achieved competitive results. Xian et al . [ 16 ] proposed four fully convolutional neural networks based on the U-Net structure. The network uses ResNet, DenseNet, and residual attention network to replace the backbone of U-Net for classification. Although the segmentation performance is greatly improved compared with U-Net, this method still has difficulties in the segmentation of small blood vessels. Fan et al . [ 17 ] proposed a method based on FCN model for coronary artery vessel segmentation. This method removes catheters and artifacts by enhancing the vascular structure of the low-contrast contrast image and improving the segmentation effect. Li et al . [ 18 ] proposed a U-Net-based CNN network called CAU-Net. To overcome the undisclosed public data set of coronary angiography images, they established a dataset containing 538 samples. Although the method has excellent denoising and recognition capabilities, it is not very good in the segmentation of small blood vessels. Pearl et al. [ 19 ] designed a two-stage vessel extraction framework to segment coronary arteries and fundus vessels. This method is mainly composed of Vessel Specific Convolutional (VSC) blocks, Skip chain Convolutional (SC) layers, and feature map summations. VSC block and SC block for better feature learning and feature propagation, the feature map summations play an important role in extracting blood vessels. Zhou et al. [ 20 ] combined deep learning algorithms with traditional algorithms and used ResNet and U-Net to segment small data sets. This algorithm has lower complexity and fewer training calculations while ensuring the segmentation effect, but it is easier to make mistakes near the end RCA bifurcation. Yang et al. [ 21 ] proposed a U-Net-based fully convolutional neural network method to segment the main blood vessels of coronary angiography images and improved the segmentation performance by quoting an improved loss function, but the network is only suitable for the segmentation of single blood vessels. Jun et al. [ 22 ] improved the U-Net-based codec structure. Unlike U-Net that has only one level skip connection between codec blocks, T-Net arranges pooling and upsampling appropriately during the coding and decoding process. All feature maps in the coding block can be connected with the decoding block. However, compared with the segmentation performance on RCA, the separation effect of this algorithm on the blood vessels on LAD and LCX still needs to be improved. Although the above methods have made great progress, there are still areas for improvement: (1) Many other tissues and noises in the background cannot be accurately segmented out, and further fine segmentation is needed for post-processing. (2) The blood vessels of other branches of the heart in the background are easily segmented. (3) The accuracy of small vessel segmentation near distal endings needs to be improved. 3 Methodology In this paper, we propose DBCU-Net network for segmentation of blood vessels in response to problems such as other networks presenting insufficient features of blood vessels, and a preprocessing method for dividing the region of interest of blood vessels in response to the problem that other background noises are easily missegmented. First, the blood vessel labeling data is divided into regions of interest, and a mask dataset is generated. Then, data enhancement and contrast enhancement operations are performed. The learning of redundant information is avoided through densely connected blocks, and the bidirectional convLSTM ensures that both forward and backward passes are used simultaneously. The two features of the encoding layer and the decoding layer are combined in each direction to realize the segmentation of blood vessels. 3.1 Data Preprocessing 3.1.1 Preprocessing of The Region of Interest Many segmentation methods for fundus vessel have been proposed in the research community, along with the publicly available dataset. In the publicly available fundus vessel images, the region of interest in the mask image is a circular white area that roughly contains the fundus vessels. Inspired by the mask images in the fundus vessel dataset, this paper uses preprocessing to process the annotated images in the coronary angiography vessel dataset. Then the region of interest generates a new mask image dataset where blood vessels exist in the images. Preprocessing effects are shown in Fig. 1 . 3.2 DBCU-Net Segmentation Algorithm 3.2.1 Network Framework In this paper, we present a neural network named DBCU-Net, as is shown in Fig. 2 . The encoding path consists of a total of three layers, each layer consists of dense connectivity modules followed by 2×2 max pooling. The output of the first convolutional la yer in each block is added to the batch normalized output, and then the feature map size is reduced to half of the original size using 2×2 max pooling. Before the max pooling is performed, the feature maps are combined by skip connection via a bidirectional BconvLSTM. Then the resulting feature maps are combined with feature maps from upsampling layer. The decoding part of DBCU-Net also has three layers, where each layer includes an upsampling layer, a bidirectional BconvLSTM module and a convolution module. The connecting layer of the encoding and decoding paths also has three layers, each with two consecutive convolution operations. We use the 2×2 upsampling operation after this and use its output as the decoding path input of the first BconvLSTM. 3.2.2 Dense Connectivity Mechanism Inspired by U-Net, the coding path proposed in this paper has four layers. Each layer of traditional U-Net consists of two consecutive 2×2 convolution and one pooling operation. As the network deepens, this series of convolutional layers learn many redundant features while helping the network to learn different types of features. To solve this problem, we apply dense connected convolution in the coding path. The feature maps of different layers in DenseNet need to be kept in the same size for the feature map connection operation. However, the downsampling layer attempts to reduce the size of the feature. In order to avoid affecting the downsampling operation in the coding path, we divide DenseNet into dense blocks and the structure is shown in Fig. 3 . 3.2.3 Bidirectional BconvLSTM Mechanism LSTM (Long Short Term Memory Network) is proposed to solve the problem that RNNs cannot handle long time dependencies by adding the state c. The state allows network to preserve long time states. The structure of LSTM is shown in Fig. 4 . The mathematical formula of LSTM is as follows: where \({h}_{t-1}\) , \({c}_{t-1}\) and \({f}_{t}\) are inputs, which are the LSTM output at the previous moment. The cell state at the previous moment and the input value of the network at the current moment, \({h}_{t}\) and \({c}_{t}\) are outputs, which are the LSTM output at the current moment and the cell state at the current moment. For the rest notations, \({i}_{t}\) is the input gate, \({f}_{t}\) is the forgetting gate, \({o}_{t}\) is the output gate, \({\tilde{c}}_{t}\) is the cell state at the current input, σ is the sigmoid function and tanh is the tanh function. In the above equations, \(W\) represents the weight matrix of each gate, \([{h}_{t-1}, {x}_{t}]\) is the two input vectors connected into a longer vector, \(b\) is the bias term of each gate, and ○ represents the corresponding multiplication by elements. The core part of LSTM is the cell state, which preserves the long-term dependency state and solves the problem that RNN cannot handle the long-term dependency. However, the LSTM uses full connectivity in the input-to-state and state-to-state transitions, which does not take into account spatial correlation and has limited ability to portray spatial features. Some scholars proposed the convLSTM [ 23 ]. The core essence of the convLSTM is the same as the LSTM, only with the addition of the convolution operation. The convolution operation can not only get the temporal relationships, but also extract spatial features. In the standard U-Net, the feature maps in the encoding path are cropped and copied to the decoding path, and then simply concatenated with the output of the corresponding upsampling layer. We propose a bidirectional convLSTM instead of simple concatenation to combine the two feature maps to obtain more accurate outputs. Traditional convLSTM only takes into account the forward information processing. However, all information in the sequence should be fully considered. To solve this problem, we use a bidirectional convLSTM in skip concatenation, which considers not only the forward dependencies but also the backward dependencies, and the structure is shown in Fig. 5 . In this paper, the output features cropped and amplified from the coding layer are denoted as \({X}_{e}\in {R}^{{C}_{l}\times {W}_{l}\times {H}_{l}}\) , and the features of output from the previous convolutional layer are denoted as \({X}_{d}\in {R}^{{C}_{l+1}\times {W}_{l+1}\times {H}_{l+1}}\) , where \({C}_{l}\) and \({W}_{l}\times {H}_{l}\) represent the number of feature channels and feature map size of the \(\text{l}\) layer features, respectively. Since each convolutional layer operation is immediately followed by a pooling operation in the coding path, \({C}_{l+1}=2\times {C}_{l}\) , \({W}_{l+1}=1/2\times {W}_{l}\) and \({H}_{l+1}=1/2\times {H}_{l}\) can be obtained. as shown in Fig. 5 , \({X}_{d}\) is first passed to the upsampling convolutional layer, in which the upsampling function and a \(2\times 2\) convolution immediately following. And after such an operation, the size of each feature map is doubled and the number of feature channels is halved, thus obtaining \({X}_{d}^{up}\in {R}^{{C}_{l}\times {W}_{l}\times {H}_{l}}\) ; then \({\widehat{X}}_{d}^{up}\) is obtained after a BN operation. then the \({X}_{e}\) and \({\widehat{X}}_{d}^{up}\) are then fed into a bidirectional convLSTM for encoding. The bidirectional convLSTM uses two convLSTMs to process the input data in two directions, forward path and backward path, and then the current input is decided by processing the dependencies of these two directions. It has been confirmed in the literature that considering bidirectional relations has better experimental performance. The convLSTM in the forward dependence and in the backward dependence can be viewed as two independent convLSTM, whose mathematical formulation is shown below. 4 Experiments And Results We train DBCU-Net on our private coronary angiography image dataset. Our dataset consists of 50 X-ray coronary angiography images along with segmentation annotation of vessel regions. Angiographic images are grayscale images of different blood vessels with a size of 512×512 pixels. Segmentation labels were annotated manually by an experienced cardiovascular clinical expert. The annotated image is a binary image with the size of 512×512 pixels corresponding to the angiographic vessel image and consists of two categories: black represents the background, and white represents the vessel. We randomly divide the 30 angiographic images and their corresponding annotated images in our dataset into the training set and divide the remaining 20 images into the testing set. Example images of original image, label (annotated image) and mask are presented in Fig. 1 . Since the coronary angiography image dataset only includes 30 images. Lack of dataset causes overfitting, and we experienced it as expected initially. To alleviate the over-fitting problem, we use data augmentation to expand the dataset so that the enhanced dataset can represent a more comprehensive dataset. To increase the robustness of our method, we adopt the following enhancement methods to the dataset: Rotation: Rotate the image on a 45-degree axis, no need to remove black edges. Mirroring: Perform horizontal mirroring and vertical mirroring on the x and y axis respectively of each group of images in the dataset. Noise injection: To study the robustness and generalization of the model, Gaussian noise with a mean value of 0 and a variance of 0.001 and salt and pepper noise with a noise ratio of 0.02 is added to the original images in the dataset. The use of Gaussian noise with zero-mean characteristics can reduce the influence of high-frequency features on the model. The 30 angiography images of the original training dataset are increased to 130 contrast images after the above-mentioned series of rotation, mirroring, and noise processing. Example images of original image, annotated image, mask, and corresponding augmented images are presented in Fig. 6 . In this experiment, we use batch training to train the network. For each angiography image after data augmentation, 3000 48x48 image patches are randomly cut out through the sliding window method and input into the network for training. In each epoch of training, we use 90% of the image patches for training and 10% for validation. Our solution is written in Tensorflow. As the experimental course involves the sensitivity of the medical field, the dataset and related code of this experiment will not be publicly processed. In our training procedure, the use of mixed precision not only speeds up the training of the network but also reduces the consumption of GPU memory. To achieve this, our Tensorflow version should be 1.14 and above. Our experiment is running on a server with 64G memory, i7-6850k CPU @3.6GHz, and the graphics card is two NVIDIA GeForce GTX 1080Ti. Our experiment is trained for 500 epochs with batch size 32 using the Adam optimizer [ 24 ] with a learning rate of 0.001. Segmentation results are given in Fig. 7 . To more intuitively observe the difference between our method and the basic U-Net network, we show the original image, expert-annotated image, and segmented image in Figure. From the figures above, we observe that compared with the U-Net segmentation method, our proposed method can better segment the coronary artery main vessel and most of the branch vessels, and it also has a certain degree of segmentation for the small blood vessels at the distal end. Meanwhile, we also analyze the segmentation results quantitatively. We used several suitable evaluation indicators, such as accuracy, recall, precision, and F1-score. As shown in Table 1, compared with other methods, our network achieved better results on our private dataset. From the result table we can see that compared to existing methods, we obtained the best results for Accuracy, Precision, Recall and F1 score with 0.985, 0.913, 0.847 and 0.879, respectively. Table.1 Evaluation metrics (Bolded font indicates the best results) Method Accuracy Precision Recall F1 AUC U-Net 0.970 0.665 0.936 0.777 0.992 Dense-UNet 0.968 0.802 0.898 0.843 0.992 ResU-Net 0.975 0.800 0.918 0.849 0.995 Dilated-BN-U-Net 0.978 0.836 0.889 0.856 0.994 DenseGAU-Net 0.975 0.832 0.900 0.860 0.992 Our Method 0.985 0.913 0.847 0.879 0.929 5 Discussion In this paper, the DBCU-Net model is proposed to improve the existing method over segmentation with insufficient accuracy. Data enhancement and region of interest segmentation are first performed on the data images; then the segmentation is performed using the DBCU-Net model vessels incorporating densely connected blocks and bidirectional convLSTM. From Fig. 7 , it can be concluded that. (1) The segmentation method using U-Net is shown in the red marked part of the figure, there will be a part of blood vessel disconnection, due to the simple series operation of U-Net's skip connection, which leads to incomplete extraction of context, image details cannot be extracted well, and the blood vessel cross and distal part segmentation is poor. (2) The DenseU-Net-based method fuses shallow and deep information in the skip-connection part, but fails to enhance the weight of the vascular segment part, and obtains insufficient effective information. (3) Dilated-BN-U-Net can fuse the null convolution to expand the sensory field, however, it still performs skip connection in a simple tandem manner and cannot fully utilize the original features. (4) The segmentation method using ResU-Net solves the problem of U-Net disconnection due to improving the simple tandem of skip connection into residual modules. The improved methods learn the target structure focusing on different shapes and sizes, and performing context extraction in coarser scales, thus solving the problem of U-Net disconnection, and the segmentation of distal vessels is also better than U-Net. However, over-segmentation occurs due to the redundancy of learning information phenomenon. (5) The segmentation method using DBCU-Net uses bidirectional convLSTM in the skip connection part, thus taking into account the forward and backward dependence of information, and the segmentation results of vessels are more refined, as shown in the yellow marked part of the figure, and the over-segmentation phenomenon is also greatly improved. Finally, the segmentation results of the algorithm in this chapter are compared and analyzed with other segmentation methods. In this paper, we show through experiments that DBCU-Net improves the over-segmentation problem of existing methods and also has a great improvement in different segmentation evaluation indexes. As shown in Table.1, our method achieves optimal results on parameters such as Accuracy, Precision, Recall and F1 score compared with other methods. However, due to the imbalance of positive and negative samples in the dataset, this paper does not score the best on the AUC evaluation, which points out the robustness of the model proposed in this paper still needs to be improved. The large cost of labeling CAG images has added challenges to the acquisition of datasets. In the future, we hope to focus on unsupervised segmentation [ 25 ] to reduce the difficulty of dataset acquisition. For a single CAG image to segment blood vessels, we may use CAG dynamic video in the future [ 26 ] to segment blood vessels to reduce the effects of blood vessel crossing, masking, and noise on the segmentatio 6 Conclusion We proposed a new convolutional neural network architecture, which is based on U-Net and combines dense connectivity and bidirectional convLSTM. The new neural network is referred to as DBU-Net. Our model is proposed in this paper is a way to enhance the current approach over segmentation with insufficient accuracy. After the evaluation of the ablation experiment, the segmentation results of the blood vessels in this method are more accurate. We conduct our experiment on our private dataset, and achieve average Accuracy, Precision, Recall and F1-score for coronary artery segmentation of 0.985, 0.913, 0.847 and 0.879 respectively. In the mainstream metrics, our method has achieved competitive results. Declarations Competing Interests We declare that we have no financial and personal relationships with other people or organizations that can inappropriately influence our work, there is no professional or other personal interest of any nature or kind in any product, service and/or company that could be construed as influencing the position presented in, or the review of, the manuscript entitled. Author Contributions Yuqiang Shen: Resources, Data curation, Project administration. Zhe Chen: Methodology, Formal analysis, Writing - original draft. Jijun Tong: Conceptualization, Methodology. Nan Jiang: Writing - review & editing, Visualization. Yun Ning: Software, Validation. Ethics approval This is an observational study. The data used in this article is obtained with the permission of the patients themselves. Consent to participate Informed consent was obtained from all individual participants included in the study.” Consent to publish The authors affirm that human research participants provided informed consent for publication of the images in Figure(s) 1, 6 and 7. References Sung H, Ferlay J, Siegel RL, Laversanne M, Soerjomataram I, Jemal A, Bray F (2021) Global cancer statistics 2020: GLOBOCAN estimates of incidence and mortality worldwide for 36 cancers in 185 countries. Cancer J Clin 71:209–249. https://doi.org/10.3322/caac.21660 Moayyedi PM, Lacy BE, Andrews CN, Enns RA, Howden CW, Vakil N (2017) ACG and CAG clinical guideline: management of dyspepsia. Official J Am Coll Gastroenterology| Acg 112:988–1013. 10.1038/ajg.2017.154 Dehkordi MT, Sadri S, Doosthoseini A (2011) A review of coronary vessel segmentation algorithms. J Med Signals Sens 1:49. https://doi.org/10.4103/2228-7477.83519 Sukanya A, Rajeswari R, Subramaniam Murugan K (2020) Region based coronary artery segmentation using modified Frangi's vesselness measure. Int J Imaging Syst Technol 30:716–730. https://doi.org/10.1002/ima.22412 Mabrouk S, Oueslati C, Ghorbel F (2017) Multiscale graph cuts based method for coronary artery segmentation in angiograms. Irbm 38:167–175. https://doi.org/10.1016/j.irbm.2017.04.004 Daoudi A, Mahmoudi S (2016) A fully automatic cardiac segmentation method using region growing technique[C]//Proceedings of the Mediterranean Conference on Pattern Recognition and Artificial Intelligence, : 103–108 Kerkeni A, Benabdallah A, Manzanera A, Bedoui MH (2016) A coronary artery segmentation method based on multiscale analysis and region growing. Comput Med Imaging Graph 48:49–61. https://doi.org/10.1016/j.compmedimag.2015.12.004 Ma G, Yang J, Huang Y, Zhao H (2019) A novel automatic coronary artery segmentation method based on region growing with annular and spherical sector partition. J Med Imaging Health Inf 9:148–152. https://doi.org/10.1166/jmihi.2019.2553 Huang G, Liu Z, Van Der Maaten L, Weinberger KQ (2017) Densely connected convolutional networks[C]//Proceedings of the Ieee Conference on Computer Vision and Pattern Recognition, : 4700–4708 Zheng S, Ristovski K, Farahat A, Gupta C (2017) Long short-term memory network for remaining useful life estimation[C]//2017 Ieee International Conference on Prognostics and Health Management (icphm): Ieee, : 88–95 Cui Z, Ke R, Pu Z, Wang Y (2018) Deep bidirectional and unidirectional LSTM recurrent neural network for network-wide traffic speed prediction. Arxiv Preprint Arxiv. https://doi.org/10.48550/arXiv.1801.02143 . :1801.02143 Ciresan D, Giusti A, Gambardella L, Schmidhuber J (2012) Deep neural networks segment neuronal membranes in electron microscopy images. Adv Neural Inf Process Syst 25. https://dl.acm.org/doi/10.5555/2999325.2999452 Ronneberger O, Fischer P, Brox T (2015) U-net: Convolutional networks for biomedical image segmentation[C]//International Conference on Medical Image Computing and Computer-assisted Intervention: Springer, Cham, : 234–241 Blaiech AG, Mansour A, Kerkeni A, Bedoui MH, Ben Abdallah A (2019) Impact of enhancement for coronary artery segmentation based on deep learning neural network[C]//Iberian Conference on Pattern Recognition and Image Analysis: Springer, Cham, : 260–272 Shi X, Du T, Chen S, Zhang H, Guan C, Xu B (2020) UENet: a novel generative adversarial network for angiography image segmentation[C]//2020 42nd Annual International Conference of the Ieee Engineering in Medicine & Biology Society (embc): Ieee, : 1612–1615 Xian Z, Wang X, Yan S, Yang D, Chen J, Peng C (2020) Main coronary vessel segmentation using deep learning in smart medical. Mathematical Problems in Engineering 2020. https://doi.org/10.1155/2020/8858344 Fan J, Du C, Song S, Cong W, Hao A, Yang J (2019) Enhanced subtraction image guided convolutional neural network for coronary artery segmentation[C]//Chinese Conference on Image and Graphics Technologies: Springer, : 625–632 Li R, Bian G, Zhou X, Xie X, Ni Z, Hou Z (2020) CAU-net: A Novel Convolutional Neural Network for Coronary Artery Segmentation in Digital Substraction Angiography[C]//International Conference on Neural Information Processing: Springer, Cham, : 185–196 Samuel PM, Veeramalai T (2021) VSSC Net: vessel specific skip chain convolutional network for blood vessel segmentation. Comput Methods Programs Biomed 198:105769. https://doi.org/10.1016/j.cmpb.2020.105769 Zhou C, Dinh TV, Kong H, Yap J, Yeo KK, Lee HK, Liang K (2021) Automated deep learning analysis of angiography video sequences for coronary artery disease. Arxiv Preprint Arxiv. https://doi.org/10.48550/arXiv.2101.12505 . :2101.12505 Yang S, Kweon J, Kim Y (2019) Major vessel segmentation on x-ray coronary angiography using deep networks with a novel penalty loss function[C]//International Conference on Medical Imaging with Deep Learning–extended Abstract Track, Jun TJ, Kweon J, Kim Y, Kim D (2020) T-net: Nested encoder–decoder architecture for the main vessel segmentation in coronary angiography. Neural Netw 128:216–233. https://doi.org/10.1016/j.neunet.2020.05.002 Song H, Wang W, Zhao S, Shen J, Lam K (2018) Pyramid dilated deeper convlstm for video salient object detection[C]//Proceedings of the European Conference on Computer Vision (eccv): Springer, Cham, : 715–731 Zhang Z (2018) Improved adam optimizer for deep neural networks[C]//2018 Ieee/acm 26th International Symposium on Quality of Service (iwqos): Springer International Publishing, : 1–2 Yu F, Zhao J, Gong Y, Wang Z, Li Y, Yang F, Dong B, Li Q, Zhang L (2019) Annotation-free cardiac vessel segmentation via knowledge transfer from retinal images[C]//International Conference on Medical Image Computing and Computer-assisted Intervention: Springer, : 714–722 Zhang J, Wang G, Xie H, Zhang S, Huang N, Zhang S, Gu L (2020) Weakly supervised vessel segmentation in X-ray angiograms by self-paced learning from noisy labels with suggestive annotation. Neurocomputing 417:114–127. https://doi.org/10.1016/j.neuc Additional Declarations No competing interests reported. Cite Share Download PDF Status: Published Journal Publication published 05 Apr, 2023 Read the published version in The International Journal of Cardiovascular Imaging → Version 1 posted Editorial decision: Major revision 14 Feb, 2023 Reviews received at journal 28 Jan, 2023 Reviewers agreed at journal 03 Jan, 2023 Reviewers agreed at journal 12 Dec, 2022 Reviewers agreed at journal 02 Oct, 2022 Reviews received at journal 13 Sep, 2022 Reviewers agreed at journal 05 Sep, 2022 Reviewers invited by journal 03 Sep, 2022 Editor assigned by journal 03 Sep, 2022 Submission checks completed at journal 03 Sep, 2022 First submitted to journal 03 Sep, 2022 You are reading this latest preprint version Research Square lets you share your work early, gain feedback from the community, and start making changes to your manuscript prior to peer review in a journal. As a division of Research Square Company, we’re committed to making research communication faster, fairer, and more useful. We do this by developing innovative software and high quality services for the global research community. Our growing team is made up of researchers and industry professionals working together to solve the most critical problems facing scientific publishing. Also discoverable on Platform About Our Team In Review Editorial Policies Advisory Board Help Center Resources Author Services Accessibility API Access RSS feed Manage Cookie Preferences © Research Square 2026 | ISSN 2693-5015 (online) Privacy Policy Terms of Service Do Not Sell My Personal Information {"props":{"pageProps":{"initialData":{"identity":"rs-2028133","acceptedTermsAndConditions":true,"allowDirectSubmit":false,"archivedVersions":[],"articleType":"Research Article","associatedPublications":[],"authors":[{"id":133879546,"identity":"8bc499a7-7385-4021-9b10-d16cf910e278","order_by":0,"name":"Yuqiang Shen","email":"","orcid":"","institution":"The Fourth Affiliated Hospital, Zhejiang University School of Medicine","correspondingAuthor":false,"submittingAuthor":false,"prefix":"","firstName":"Yuqiang","middleName":"","lastName":"Shen","suffix":""},{"id":133879547,"identity":"59628864-544f-40cf-9bbf-7e5cc1e40c6d","order_by":1,"name":"Zhe Chen","email":"","orcid":"","institution":"Zhejiang Sci-Tech University","correspondingAuthor":false,"submittingAuthor":false,"prefix":"","firstName":"Zhe","middleName":"","lastName":"Chen","suffix":""},{"id":133879548,"identity":"82d718b1-e3fd-48ac-88db-4a5c0ef39a07","order_by":2,"name":"Jijun Tong","email":"","orcid":"","institution":"Zhejiang Sci-Tech University","correspondingAuthor":false,"submittingAuthor":false,"prefix":"","firstName":"Jijun","middleName":"","lastName":"Tong","suffix":""},{"id":133879549,"identity":"ccb033f0-2fed-47b8-bfe8-7c6e46c29aa9","order_by":3,"name":"Nan Jiang","email":"data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAAZAAAAAyAQMAAABI0h/eAAAABlBMVEX///8AAABVwtN+AAAACXBIWXMAAA7EAAAOxAGVKw4bAAAAzElEQVRIiWNgGAWjYBACxmYwZQPlshGvJQ2qmhgtUHCYBC3M7czPHn7dcd5efn6PAcOHssMM/LMbCDmMzdxY9sxtZoNjPAaMM84dZpC4c4CQFgYzacm222wGbDwGzLxthxkMJBIIaWH/BtRyjke+DajlL3FaeMwkP7YdkGAAOoyZkUgtZdKMbckGBsfSCg72nEvnkbhBQIth//Ftkj/b7Ozlmw9vfPCjzFqOfwYhLQ3AgOaBcg4AMQ9utVAgD3LcD4LKRsEoGAWjYEQDAFWfOqYqK0B9AAAAAElFTkSuQmCC","orcid":"","institution":"Zhejiang Sci-Tech University","correspondingAuthor":true,"submittingAuthor":false,"prefix":"","firstName":"Nan","middleName":"","lastName":"Jiang","suffix":""},{"id":133879550,"identity":"873268b5-4e2e-4f3d-a8c7-b93e3c7cc72d","order_by":4,"name":"Yun Ning","email":"","orcid":"","institution":"Zhejiang Sci-Tech University","correspondingAuthor":false,"submittingAuthor":false,"prefix":"","firstName":"Yun","middleName":"","lastName":"Ning","suffix":""}],"badges":[],"createdAt":"2022-09-03 09:14:19","currentVersionCode":1,"declarations":"","doi":"10.21203/rs.3.rs-2028133/v1","doiUrl":"https://doi.org/10.21203/rs.3.rs-2028133/v1","draftVersion":[],"editorialEvents":[{"content":"https://doi.org/10.1007/s10554-023-02849-3","type":"published","date":"2023-04-05T20:22:43+00:00"}],"editorialNote":"","failedWorkflow":false,"files":[{"id":26180646,"identity":"b11f68bd-3dc9-4acb-b470-a4bff6c181f4","added_by":"auto","created_at":"2022-09-07 14:45:32","extension":"png","order_by":1,"title":"Figure 1","display":"","copyAsset":false,"role":"figure","size":156800,"visible":true,"origin":"","legend":"\u003cp\u003eSchematic diagram of the region of interest processing.\u003c/p\u003e\u003cp\u003e\u003cbr\u003e\u003c/p\u003e","description":"","filename":"Fig1.png","url":"https://assets-eu.researchsquare.com/files/rs-2028133/v1/3d91c8941ea1b756f20301b1.png"},{"id":26180645,"identity":"4ee0451e-b148-4092-ad95-1def84e83e37","added_by":"auto","created_at":"2022-09-07 14:45:32","extension":"png","order_by":2,"title":"Figure 2","display":"","copyAsset":false,"role":"figure","size":2378007,"visible":true,"origin":"","legend":"\u003cp\u003eThe overall architecture of DBCU-Net.\u003c/p\u003e\u003cp\u003e\u003cbr\u003e\u003c/p\u003e","description":"","filename":"Fig2.png","url":"https://assets-eu.researchsquare.com/files/rs-2028133/v1/6189c63323346633d57aa4cd.png"},{"id":26181378,"identity":"c0cb7a67-df54-4d19-84ee-e366d0390a3d","added_by":"auto","created_at":"2022-09-07 14:55:32","extension":"png","order_by":3,"title":"Figure 3","display":"","copyAsset":false,"role":"figure","size":21474,"visible":true,"origin":"","legend":"\u003cp\u003eDense blocks in encoding part and decoding part.\u003c/p\u003e\u003cp\u003e\u003cbr\u003e\u003c/p\u003e","description":"","filename":"Fig3.png","url":"https://assets-eu.researchsquare.com/files/rs-2028133/v1/9855fd5440e8462f66d0a3db.png"},{"id":26181053,"identity":"9575344a-9ae6-460f-934b-d1432c37a106","added_by":"auto","created_at":"2022-09-07 14:50:32","extension":"png","order_by":4,"title":"Figure 4","display":"","copyAsset":false,"role":"figure","size":19672,"visible":true,"origin":"","legend":"\u003cp\u003eThe details of LSTM Architecture\u003c/p\u003e\u003cp\u003e\u003cbr\u003e\u003c/p\u003e","description":"","filename":"Fig4.png","url":"https://assets-eu.researchsquare.com/files/rs-2028133/v1/5a5af02604d4880bb48474f4.png"},{"id":26180648,"identity":"0440bfd1-f677-41eb-af1b-effd60f55edf","added_by":"auto","created_at":"2022-09-07 14:45:32","extension":"png","order_by":5,"title":"Figure 5","display":"","copyAsset":false,"role":"figure","size":45063,"visible":true,"origin":"","legend":"\u003cp\u003eBidirectional convLSTM Architecture\u003c/p\u003e\u003cp\u003e\u003cbr\u003e\u003c/p\u003e","description":"","filename":"Fig5.png","url":"https://assets-eu.researchsquare.com/files/rs-2028133/v1/81a243fc0d488647540da243.png"},{"id":26180651,"identity":"ed76e7bf-9e42-47d9-86be-366a5f0dc5f2","added_by":"auto","created_at":"2022-09-07 14:45:33","extension":"png","order_by":6,"title":"Figure 6","display":"","copyAsset":false,"role":"figure","size":287327,"visible":true,"origin":"","legend":"\u003cp\u003eAugmented images of CAG\u003c/p\u003e\u003cp\u003e\u003cbr\u003e\u003c/p\u003e","description":"","filename":"Fig6.png","url":"https://assets-eu.researchsquare.com/files/rs-2028133/v1/1843c7ebdeea4034bd0d92ee.png"},{"id":26181051,"identity":"5ca36269-1494-48aa-aa11-6e9cfe144ba3","added_by":"auto","created_at":"2022-09-07 14:50:32","extension":"png","order_by":7,"title":"Figure 7","display":"","copyAsset":false,"role":"figure","size":310522,"visible":true,"origin":"","legend":"\u003cp\u003eSegmentation results: a. Original image; b. Annotated image; c. U-Net; d. DenseU-Net;\u0026nbsp;e. ResU-Net; f. Dilated-bn-Net; g. DenseGAU-Net; h. DBCU- Net;\u003c/p\u003e\u003cp\u003e\u003cbr\u003e\u003c/p\u003e","description":"","filename":"Fig7.png","url":"https://assets-eu.researchsquare.com/files/rs-2028133/v1/2a16d4f07093e14e1b4fee27.png"},{"id":44724255,"identity":"9bacb7cd-80f2-4c4e-86d8-6e00a0391970","added_by":"auto","created_at":"2023-10-16 20:28:06","extension":"pdf","order_by":0,"title":"","display":"","copyAsset":false,"role":"manuscript-pdf","size":1358490,"visible":true,"origin":"","legend":"","description":"","filename":"manuscript.pdf","url":"https://assets-eu.researchsquare.com/files/rs-2028133/v1/eadc9c84-de0c-46c1-a734-0cd6eb1a231e.pdf"}],"financialInterests":"No competing interests reported.","formattedTitle":"DBCU-Net: Deep Learning Approach for Segmentation of Coronary Angiography Images","fulltext":[{"header":"1 Introduction","content":"\u003cp\u003eCoronary artery disease (CAD) is a common heart disease, which is the leading cause of death worldwide according to the World Health Organization (WHO) [\u003cspan citationid=\"CR1\" class=\"CitationRef\"\u003e1\u003c/span\u003e]. Globally, the number of patients with CAD is expected to increase from 327.9\u0026nbsp;million in 2017 to 365.9\u0026nbsp;million in 2026. CAD poses a huge threat to human life. An accurate diagnosis of CAD is particularly important. With the development of computer diagnosis and treatment technologies, more powerful medical methods, such as magnetic resonance imaging(MRI), computed tomography(CT) [\u003cspan citationid=\"CR2\" class=\"CitationRef\"\u003e2\u003c/span\u003e], and X-ray coronary angiography(CAG)[\u003cspan citationid=\"CR3\" class=\"CitationRef\"\u003e3\u003c/span\u003e], are proposed to assist diagnosis. As the \"gold standard\" for the diagnosis of CAD, CAG can precisely pinpoint the site and the degree of coronary artery stenosis, as well as the symptoms of the condition.\u003c/p\u003e \u003cp\u003eAccurate medical image analysis is crucial to subsequent clinical diagnosis and treatment. Diagnosis and treatment of clinical diseases mainly rely on advanced instruments and doctors' high-precision technology. However, manual segmentation of such CAG images requires a lot of medical expertise, which is time-consuming and is prone to human error. Due to the shortage of medical resources, a computer diagnosis system that can assist doctors in making better diagnoses and determining follow-up treatment plans in a shorter time is preferred.\u003c/p\u003e \u003cp\u003eIn normal CAG image segmentation, the images are divided into two parts: vessels and backgrounds. Although traditional machine learning techniques (e.g. model-based methods and atlas-based methods [\u003cspan citationid=\"CR4\" class=\"CitationRef\"\u003e4\u003c/span\u003e, \u003cspan citationid=\"CR5\" class=\"CitationRef\"\u003e5\u003c/span\u003e, \u003cspan citationid=\"CR6\" class=\"CitationRef\"\u003e6\u003c/span\u003e, \u003cspan citationid=\"CR7\" class=\"CitationRef\"\u003e7\u003c/span\u003e, \u003cspan citationid=\"CR8\" class=\"CitationRef\"\u003e8\u003c/span\u003e]) have achieved good performance in vessels segmentation, deep learning methods have defeated traditional techniques with its automation and versatility in terms of segmentation efficiency and detection accuracy.\u003c/p\u003e \u003cp\u003eIn this paper, we propose a DBCU-Net for automatic segmentation of CAG images, which is an extension of \u003cspan type=\"Underline\" class=\"Underline\" name=\"Emphasis\"\u003eU-Net\u003c/span\u003e, \u003cspan type=\"Underline\" class=\"Underline\" name=\"Emphasis\"\u003eD\u003c/span\u003eenseNet [\u003cspan citationid=\"CR9\" class=\"CitationRef\"\u003e9\u003c/span\u003e] with \u003cspan type=\"Underline\" class=\"Underline\" name=\"Emphasis\"\u003eb\u003c/span\u003eidirectional \u003cspan type=\"Underline\" class=\"Underline\" name=\"Emphasis\"\u003ec\u003c/span\u003eonvLSTM [\u003cspan citationid=\"CR10\" class=\"CitationRef\"\u003e10\u003c/span\u003e, \u003cspan citationid=\"CR11\" class=\"CitationRef\"\u003e11\u003c/span\u003e]. The dense connectivity strengthens feature propagation and gets multi-level feature to enhance representation. The bidirectional convLSTM ensures both forward and backward passes are used simultaneously. The two features of the encoding layer and the decoding layer are combined in each direction to realize the segmentation of blood vessels.\u003c/p\u003e \u003cp\u003eThe contributions of our works are listed as follows:\u003c/p\u003e \u003cp\u003e(1) Many other tissues and noises in the background can be accurately segmented.\u003c/p\u003e \u003cp\u003e(2) The problem of missegmentation of vessels in other branches of the heart in the background is improved.\u003c/p\u003e \u003cp\u003e(3) The segmentation of tiny vessels in the distal terminals has the same positive outcomes. We conduct our experiment on our private dataset, and obtain competitive performance.\u003c/p\u003e"},{"header":"2 Related Work","content":"\u003cp\u003eSince 2010, deep learning has outperformed the conventional state-of-the-art approaches towards visual recognition tasks like heart MRI segmentation and cancer lung nodule segmentation. Ciresan \u003cem\u003eet al\u003c/em\u003e [\u003cspan citationid=\"CR12\" class=\"CitationRef\"\u003e12\u003c/span\u003e]. used CNN to automatically segment electron microscope images for the first time, and won the electron microscope challenge with a huge advantage. In 2015, Ronneberger \u003cem\u003eet al\u003c/em\u003e [\u003cspan citationid=\"CR13\" class=\"CitationRef\"\u003e13\u003c/span\u003e]. proposed a groundbreaking network architecture named U-Net. U-Net is composed of an encoder-decoder structure, where the encoding path includes four down-sampling layers, a total of 16 times down-sampling. Symmetrically, its decoding path is up-sampled four times accordingly. The high-level semantic information feature map obtained from the encoding path is restored to the resolution of the original image, and the feature map stitching is performed on the same stage using skipping connection. Finally, a 1x1 convolution is used to obtain the required number of classes (e.g. vessels and background) from each component feature.\u003c/p\u003e \u003cp\u003eAlthough U-Net has achieved good results in medical image segmentation, it also has some limitations. First of all, a large and deep network requires about 19\u0026nbsp;million parameters. To better match the network, the quantity of date needed for training is large. Secondly, to make the network achieve higher accuracy, more convolutions need to be added to the network, which causes the network has to learn more unnecessary information that leads to overfitting. Finally, as the network deepens, there may be a risk of gradient disappearance, which brings great risks to the training of deep neural networks.\u003c/p\u003e \u003cp\u003eTo solve the above-mentioned architectural problems, many U-Net-based network deformation structures have been proposed and achieved good results. A new U-Net-based DNN structure was proposed to improve the segmentation effect by Ahmed \u003cem\u003eet al\u003c/em\u003e. [\u003cspan citationid=\"CR14\" class=\"CitationRef\"\u003e14\u003c/span\u003e] This method selected various enhancement methods (e.g. histogram and Frangi multi-scale filtering) according to the background to remove noise and improve segment effect. Shi \u003cem\u003eet al\u003c/em\u003e. [\u003cspan citationid=\"CR15\" class=\"CitationRef\"\u003e15\u003c/span\u003e] proposed an adaptive generative confrontation network to complete the segmentation of coronary blood vessels. This method uses an adaptive U network as a generator and a three-layer pyramid structure as a discriminator. Under the mechanism of generating confrontation, both the generators can extract the fine features of coronary arteries. Compared with other blood vessel segmentation methods, the segmentation accuracy and continuity of this method have been greatly improved. However, for U-Net, this method requires multiple prior information.\u003c/p\u003e \u003cp\u003eWith the continuous development of deep learning in medical image segmentation, many scholars have combined U-Net with other deep neural networks and achieved competitive results. Xian \u003cem\u003eet al\u003c/em\u003e. [\u003cspan citationid=\"CR16\" class=\"CitationRef\"\u003e16\u003c/span\u003e] proposed four fully convolutional neural networks based on the U-Net structure. The network uses ResNet, DenseNet, and residual attention network to replace the backbone of U-Net for classification. Although the segmentation performance is greatly improved compared with U-Net, this method still has difficulties in the segmentation of small blood vessels. Fan \u003cem\u003eet al\u003c/em\u003e. [\u003cspan citationid=\"CR17\" class=\"CitationRef\"\u003e17\u003c/span\u003e] proposed a method based on FCN model for coronary artery vessel segmentation. This method removes catheters and artifacts by enhancing the vascular structure of the low-contrast contrast image and improving the segmentation effect. Li \u003cem\u003eet al\u003c/em\u003e. [\u003cspan citationid=\"CR18\" class=\"CitationRef\"\u003e18\u003c/span\u003e] proposed a U-Net-based CNN network called CAU-Net. To overcome the undisclosed public data set of coronary angiography images, they established a dataset containing 538 samples. Although the method has excellent denoising and recognition capabilities, it is not very good in the segmentation of small blood vessels. Pearl \u003cem\u003eet al.\u003c/em\u003e [\u003cspan citationid=\"CR19\" class=\"CitationRef\"\u003e19\u003c/span\u003e] designed a two-stage vessel extraction framework to segment coronary arteries and fundus vessels. This method is mainly composed of Vessel Specific Convolutional (VSC) blocks, Skip chain Convolutional (SC) layers, and feature map summations. VSC block and SC block for better feature learning and feature propagation, the feature map summations play an important role in extracting blood vessels. Zhou \u003cem\u003eet al.\u003c/em\u003e [\u003cspan citationid=\"CR20\" class=\"CitationRef\"\u003e20\u003c/span\u003e] combined deep learning algorithms with traditional algorithms and used ResNet and U-Net to segment small data sets. This algorithm has lower complexity and fewer training calculations while ensuring the segmentation effect, but it is easier to make mistakes near the end RCA bifurcation. Yang \u003cem\u003eet al.\u003c/em\u003e [\u003cspan citationid=\"CR21\" class=\"CitationRef\"\u003e21\u003c/span\u003e] proposed a U-Net-based fully convolutional neural network method to segment the main blood vessels of coronary angiography images and improved the segmentation performance by quoting an improved loss function, but the network is only suitable for the segmentation of single blood vessels. Jun \u003cem\u003eet al.\u003c/em\u003e [\u003cspan citationid=\"CR22\" class=\"CitationRef\"\u003e22\u003c/span\u003e] improved the U-Net-based codec structure. Unlike U-Net that has only one level skip connection between codec blocks, T-Net arranges pooling and upsampling appropriately during the coding and decoding process. All feature maps in the coding block can be connected with the decoding block. However, compared with the segmentation performance on RCA, the separation effect of this algorithm on the blood vessels on LAD and LCX still needs to be improved.\u003c/p\u003e \u003cp\u003eAlthough the above methods have made great progress, there are still areas for improvement:\u003c/p\u003e \u003cp\u003e(1) Many other tissues and noises in the background cannot be accurately segmented out, and further fine segmentation is needed for post-processing.\u003c/p\u003e \u003cp\u003e(2) The blood vessels of other branches of the heart in the background are easily segmented.\u003c/p\u003e \u003cp\u003e(3) The accuracy of small vessel segmentation near distal endings needs to be improved.\u003c/p\u003e"},{"header":"3 Methodology","content":"\u003cp\u003eIn this paper, we propose DBCU-Net network for segmentation of blood vessels in response to problems such as other networks presenting insufficient features of blood vessels, and a preprocessing method for dividing the region of interest of blood vessels in response to the problem that other background noises are easily missegmented. First, the blood vessel labeling data is divided into regions of interest, and a mask dataset is generated. Then, data enhancement and contrast enhancement operations are performed. The learning of redundant information is avoided through densely connected blocks, and the bidirectional convLSTM ensures that both forward and backward passes are used simultaneously. The two features of the encoding layer and the decoding layer are combined in each direction to realize the segmentation of blood vessels.\u003c/p\u003e\n\u003cdiv class=\"Section2\" id=\"Sec4\"\u003e\n \u003ch2\u003e3.1 Data Preprocessing\u003c/h2\u003e\n \u003cdiv class=\"Section3\" id=\"Sec5\"\u003e\n \u003ch2\u003e3.1.1 Preprocessing of The Region of Interest\u003c/h2\u003e\n \u003cp\u003eMany segmentation methods for fundus vessel have been proposed in the research community, along with the publicly available dataset. In the publicly available fundus vessel images, the region of interest in the mask image is a circular white area that roughly contains the fundus vessels. Inspired by the mask images in the fundus vessel dataset, this paper uses preprocessing to process the annotated images in the coronary angiography vessel dataset. Then the region of interest generates a new mask image dataset where blood vessels exist in the images. Preprocessing effects are shown in Fig. \u003cspan class=\"InternalRef\"\u003e1\u003c/span\u003e.\u003c/p\u003e\n \u003c/div\u003e\n\u003c/div\u003e\n\u003cdiv class=\"Section2\" id=\"Sec6\"\u003e\n \u003ch2\u003e3.2 DBCU-Net Segmentation Algorithm\u003c/h2\u003e\n \u003cdiv class=\"Section3\" id=\"Sec7\"\u003e\n \u003ch2\u003e3.2.1 Network Framework\u003c/h2\u003e\n \u003cp\u003eIn this paper, we present a neural network named DBCU-Net, as is shown in Fig. \u003cspan class=\"InternalRef\"\u003e2\u003c/span\u003e. The encoding path consists of a total of three layers, each layer consists of dense connectivity modules followed by 2\u0026times;2 max pooling.\u003c/p\u003e\n \u003cp\u003eThe output of the first convolutional la yer in each block is added to the batch normalized output, and then the feature map size is reduced to half of the original size using 2\u0026times;2 max pooling. Before the max pooling is performed, the feature maps are combined by skip connection via a bidirectional BconvLSTM. Then the resulting feature maps are combined with feature maps from upsampling layer.\u003c/p\u003e\n \u003cp\u003eThe decoding part of DBCU-Net also has three layers, where each layer includes an upsampling layer, a bidirectional BconvLSTM module and a convolution module. The connecting layer of the encoding and decoding paths also has three layers, each with two consecutive convolution operations. We use the 2\u0026times;2 upsampling operation after this and use its output as the decoding path input of the first BconvLSTM.\u003c/p\u003e\n \u003c/div\u003e\n \u003cdiv class=\"Section3\" id=\"Sec8\"\u003e\n \u003ch2\u003e3.2.2 Dense Connectivity Mechanism\u003c/h2\u003e\n \u003cp\u003eInspired by U-Net, the coding path proposed in this paper has four layers. Each layer of traditional U-Net consists of two consecutive 2\u0026times;2 convolution and one pooling operation. As the network deepens, this series of convolutional layers learn many redundant features while helping the network to learn different types of features. To solve this problem, we apply dense connected convolution in the coding path. The feature maps of different layers in DenseNet need to be kept in the same size for the feature map connection operation. However, the downsampling layer attempts to reduce the size of the feature. In order to avoid affecting the downsampling operation in the coding path, we divide DenseNet into dense blocks and the structure is shown in Fig. \u003cspan class=\"InternalRef\"\u003e3\u003c/span\u003e.\u003c/p\u003e\n \u003c/div\u003e\n \u003cdiv class=\"Section3\" id=\"Sec9\"\u003e\n \u003ch2\u003e3.2.3 Bidirectional BconvLSTM Mechanism\u003c/h2\u003e\n \u003cp\u003eLSTM (Long Short Term Memory Network) is proposed to solve the problem that RNNs cannot handle long time dependencies by adding the state c. The state allows network to preserve long time states. The structure of LSTM is shown in Fig. \u003cspan class=\"InternalRef\"\u003e4\u003c/span\u003e.\u003c/p\u003e\n \u003cp\u003eThe mathematical formula of LSTM is as follows:\u003c/p\u003e\n \u003cdiv class=\"Equation\" id=\"Equ1\"\u003e\n \u003cdiv class=\"mathdisplay\" id=\"FileID_Equ1\" name=\"EquationSource\"\u003e\u003cimg src=\"data:image/png;base64,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\"\u003e\u003c/div\u003e\n \u003c/div\u003e\n \u003cdiv class=\"Equation\" id=\"Equ6\"\u003ewhere \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\({h}_{t-1}\\)\u003c/span\u003e\u003c/span\u003e, \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\({c}_{t-1}\\)\u003c/span\u003e\u003c/span\u003e and \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\({f}_{t}\\)\u003c/span\u003e\u003c/span\u003e are inputs, which are the LSTM output at the previous moment. The cell state at the previous moment and the input value of the network at the current moment, \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\({h}_{t}\\)\u003c/span\u003e\u003c/span\u003e and \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\({c}_{t}\\)\u003c/span\u003e\u003c/span\u003e are outputs, which are the LSTM output at the current moment and the cell state at the current moment. For the rest notations, \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\({i}_{t}\\)\u003c/span\u003e\u003c/span\u003e is the input gate, \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\({f}_{t}\\)\u003c/span\u003e\u003c/span\u003e is the forgetting gate, \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\({o}_{t}\\)\u003c/span\u003e\u003c/span\u003e is the output gate, \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\({\\tilde{c}}_{t}\\)\u003c/span\u003e\u003c/span\u003e is the cell state at the current input, \u0026sigma; is the sigmoid function and tanh is the tanh function.\u003c/div\u003e\n \u003cp\u003eIn the above equations, \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(W\\)\u003c/span\u003e\u003c/span\u003e represents the weight matrix of each gate, \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\([{h}_{t-1}, {x}_{t}]\\)\u003c/span\u003e\u003c/span\u003e is the two input vectors connected into a longer vector, \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(b\\)\u003c/span\u003e\u003c/span\u003e is the bias term of each gate, and ○ represents the corresponding multiplication by elements.\u003c/p\u003e\n \u003cp\u003eThe core part of LSTM is the cell state, which preserves the long-term dependency state and solves the problem that RNN cannot handle the long-term dependency. However, the LSTM uses full connectivity in the input-to-state and state-to-state transitions, which does not take into account spatial correlation and has limited ability to portray spatial features. Some scholars proposed the convLSTM [\u003cspan class=\"CitationRef\"\u003e23\u003c/span\u003e]. The core essence of the convLSTM is the same as the LSTM, only with the addition of the convolution operation. The convolution operation can not only get the temporal relationships, but also extract spatial features.\u003c/p\u003e\n \u003cp\u003eIn the standard U-Net, the feature maps in the encoding path are cropped and copied to the decoding path, and then simply concatenated with the output of the corresponding upsampling layer. We propose a bidirectional convLSTM instead of simple concatenation to combine the two feature maps to obtain more accurate outputs. Traditional convLSTM only takes into account the forward information processing. However, all information in the sequence should be fully considered. To solve this problem, we use a bidirectional convLSTM in skip concatenation, which considers not only the forward dependencies but also the backward dependencies, and the structure is shown in Fig. \u003cspan class=\"InternalRef\"\u003e5\u003c/span\u003e.\u003c/p\u003e\n \u003cp\u003eIn this paper, the output features cropped and amplified from the coding layer are denoted as \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\({X}_{e}\\in {R}^{{C}_{l}\\times {W}_{l}\\times {H}_{l}}\\)\u003c/span\u003e\u003c/span\u003e, and the features of output from the previous convolutional layer are denoted as \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\({X}_{d}\\in {R}^{{C}_{l+1}\\times {W}_{l+1}\\times {H}_{l+1}}\\)\u003c/span\u003e\u003c/span\u003e, where \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\({C}_{l}\\)\u003c/span\u003e\u003c/span\u003e and \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\({W}_{l}\\times {H}_{l}\\)\u003c/span\u003e\u003c/span\u003e represent the number of feature channels and feature map size of the \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\text{l}\\)\u003c/span\u003e\u003c/span\u003e layer features, respectively. Since each convolutional layer operation is immediately followed by a pooling operation in the coding path, \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\({C}_{l+1}=2\\times {C}_{l}\\)\u003c/span\u003e\u003c/span\u003e, \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\({W}_{l+1}=1/2\\times {W}_{l}\\)\u003c/span\u003e\u003c/span\u003e and \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\({H}_{l+1}=1/2\\times {H}_{l}\\)\u003c/span\u003e\u003c/span\u003e can be obtained. as shown in Fig. \u003cspan class=\"InternalRef\"\u003e5\u003c/span\u003e, \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\({X}_{d}\\)\u003c/span\u003e\u003c/span\u003e is first passed to the upsampling convolutional layer, in which the upsampling function and a \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(2\\times 2\\)\u003c/span\u003e\u003c/span\u003e convolution immediately following. And after such an operation, the size of each feature map is doubled and the number of feature channels is halved, thus obtaining \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\({X}_{d}^{up}\\in {R}^{{C}_{l}\\times {W}_{l}\\times {H}_{l}}\\)\u003c/span\u003e\u003c/span\u003e; then \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\({\\widehat{X}}_{d}^{up}\\)\u003c/span\u003e\u003c/span\u003e is obtained after a BN operation. then the \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\({X}_{e}\\)\u003c/span\u003e\u003c/span\u003e and \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\({\\widehat{X}}_{d}^{up}\\)\u003c/span\u003e\u003c/span\u003e are then fed into a bidirectional convLSTM for encoding. The bidirectional convLSTM uses two convLSTMs to process the input data in two directions, forward path and backward path, and then the current input is decided by processing the dependencies of these two directions. It has been confirmed in the literature that considering bidirectional relations has better experimental performance. The convLSTM in the forward dependence and in the backward dependence can be viewed as two independent convLSTM, whose mathematical formulation is shown below.\u003c/p\u003e\n \u003cdiv class=\"Equation\" id=\"Equ7\"\u003e\n \u003cdiv class=\"mathdisplay\" id=\"FileID_Equ7\" name=\"EquationSource\"\u003e\u003cimg src=\"data:image/png;base64,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\"\u003e\u003c/div\u003e\n \u003c/div\u003e\n \u003cdiv class=\"Equation\" id=\"Equ12\"\u003e\u003cbr\u003e\u003c/div\u003e\n \u003c/div\u003e\n\u003c/div\u003e"},{"header":"4 Experiments And Results","content":"\u003cp\u003eWe train DBCU-Net on our private coronary angiography image dataset. Our dataset consists of 50 X-ray coronary angiography images along with segmentation annotation of vessel regions. Angiographic images are grayscale images of different blood vessels with a size of 512\u0026times;512 pixels. Segmentation labels were annotated manually by an experienced cardiovascular clinical expert. The annotated image is a binary image with the size of 512\u0026times;512 pixels corresponding to the angiographic vessel image and consists of two categories: black represents the background, and white represents the vessel. We randomly divide the 30 angiographic images and their corresponding annotated images in our dataset into the training set and divide the remaining 20 images into the testing set. Example images of original image, label (annotated image) and mask are presented in Fig.\u0026nbsp;\u003cspan refid=\"Fig1\" class=\"InternalRef\"\u003e1\u003c/span\u003e.\u003c/p\u003e \u003cp\u003eSince the coronary angiography image dataset only includes 30 images. Lack of dataset causes overfitting, and we experienced it as expected initially. To alleviate the over-fitting problem, we use data augmentation to expand the dataset so that the enhanced dataset can represent a more comprehensive dataset. To increase the robustness of our method, we adopt the following enhancement methods to the dataset:\u003c/p\u003e \u003cp\u003e \u003col\u003e \u003cspan\u003e \u003cli\u003e \u003cp\u003eRotation: Rotate the image on a 45-degree axis, no need to remove black edges.\u003c/p\u003e \u003c/li\u003e \u003c/span\u003e \u003cspan\u003e \u003cli\u003e \u003cp\u003eMirroring: Perform horizontal mirroring and vertical mirroring on the x and y axis respectively of each group of images in the dataset.\u003c/p\u003e \u003c/li\u003e \u003c/span\u003e \u003cspan\u003e \u003cli\u003e \u003cp\u003eNoise injection: To study the robustness and generalization of the model, Gaussian noise with a mean value of 0 and a variance of 0.001 and salt and pepper noise with a noise ratio of 0.02 is added to the original images in the dataset. The use of Gaussian noise with zero-mean characteristics can reduce the influence of high-frequency features on the model.\u003c/p\u003e \u003c/li\u003e \u003c/span\u003e \u003c/ol\u003e \u003c/p\u003e \u003cp\u003eThe 30 angiography images of the original training dataset are increased to 130 contrast images after the above-mentioned series of rotation, mirroring, and noise processing. Example images of original image, annotated image, mask, and corresponding augmented images are presented in Fig.\u0026nbsp;\u003cspan refid=\"Fig6\" class=\"InternalRef\"\u003e6\u003c/span\u003e. In this experiment, we use batch training to train the network. For each angiography image after data augmentation, 3000 48x48 image patches are randomly cut out through the sliding window method and input into the network for training. In each epoch of training, we use 90% of the image patches for training and 10% for validation.\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003eOur solution is written in Tensorflow. As the experimental course involves the sensitivity of the medical field, the dataset and related code of this experiment will not be publicly processed. In our training procedure, the use of mixed precision not only speeds up the training of the network but also reduces the consumption of GPU memory. To achieve this, our Tensorflow version should be 1.14 and above. Our experiment is running on a server with 64G memory, i7-6850k CPU @3.6GHz, and the graphics card is two NVIDIA GeForce GTX 1080Ti.\u003c/p\u003e \u003cp\u003eOur experiment is trained for 500 epochs with batch size 32 using the Adam optimizer [\u003cspan citationid=\"CR24\" class=\"CitationRef\"\u003e24\u003c/span\u003e] with a learning rate of 0.001. Segmentation results are given in Fig.\u0026nbsp;\u003cspan refid=\"Fig7\" class=\"InternalRef\"\u003e7\u003c/span\u003e.\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003eTo more intuitively observe the difference between our method and the basic U-Net network, we show the original image, expert-annotated image, and segmented image in Figure. From the figures above, we observe that compared with the U-Net segmentation method, our proposed method can better segment the coronary artery main vessel and most of the branch vessels, and it also has a certain degree of segmentation for the small blood vessels at the distal end. Meanwhile, we also analyze the segmentation results quantitatively. We used several suitable evaluation indicators, such as accuracy, recall, precision, and F1-score. As shown in Table\u0026nbsp;1, compared with other methods, our network achieved better results on our private dataset. From the result table we can see that compared to existing methods, we obtained the best results for Accuracy, Precision, Recall and F1 score with 0.985, 0.913, 0.847 and 0.879, respectively.\u003c/p\u003e \u003cp\u003eTable.1 Evaluation metrics (Bolded font indicates the best results)\u003c/p\u003e \u003cp\u003e \u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"No\" id=\"Taba\" border=\"1\"\u003e \u003ccolgroup cols=\"6\"\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c4\" colnum=\"4\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c5\" colnum=\"5\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c6\" colnum=\"6\"\u003e\u003c/div\u003e \u003cthead\u003e \u003ctr\u003e \u003cth align=\"left\" colname=\"c1\"\u003e \u003cp\u003eMethod\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c2\"\u003e \u003cp\u003e\u003cem\u003eAccuracy\u003c/em\u003e\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c3\"\u003e \u003cp\u003e\u003cem\u003ePrecision\u003c/em\u003e\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c4\"\u003e \u003cp\u003e\u003cem\u003eRecall\u003c/em\u003e\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c5\"\u003e \u003cp\u003e\u003cem\u003eF1\u003c/em\u003e\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c6\"\u003e \u003cp\u003e\u003cem\u003eAUC\u003c/em\u003e\u003c/p\u003e \u003c/th\u003e \u003c/tr\u003e \u003c/thead\u003e \u003ctbody\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eU-Net\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e0.970\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0.665\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.936\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0.777\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e0.992\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eDense-UNet\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e0.968\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0.802\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.898\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0.843\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e0.992\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eResU-Net\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e0.975\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0.800\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.918\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0.849\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e\u003cb\u003e0.995\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eDilated-BN-U-Net\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e0.978\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0.836\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.889\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0.856\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e0.994\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eDenseGAU-Net\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e0.975\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0.832\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.900\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0.860\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e0.992\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eOur Method\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e\u003cb\u003e0.985\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e\u003cb\u003e0.913\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e\u003cb\u003e0.847\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e\u003cb\u003e0.879\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e0.929\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003c/tbody\u003e \u003c/colgroup\u003e \u003c/table\u003e\u003c/div\u003e \u003c/p\u003e"},{"header":"5 Discussion","content":"\u003cp\u003eIn this paper, the DBCU-Net model is proposed to improve the existing method over segmentation with insufficient accuracy. Data enhancement and region of interest segmentation are first performed on the data images; then the segmentation is performed using the DBCU-Net model vessels incorporating densely connected blocks and bidirectional convLSTM.\u003c/p\u003e \u003cp\u003eFrom Fig.\u0026nbsp;\u003cspan refid=\"Fig7\" class=\"InternalRef\"\u003e7\u003c/span\u003e, it can be concluded that. (1) The segmentation method using U-Net is shown in the red marked part of the figure, there will be a part of blood vessel disconnection, due to the simple series operation of U-Net's skip connection, which leads to incomplete extraction of context, image details cannot be extracted well, and the blood vessel cross and distal part segmentation is poor. (2) The DenseU-Net-based method fuses shallow and deep information in the skip-connection part, but fails to enhance the weight of the vascular segment part, and obtains insufficient effective information. (3) Dilated-BN-U-Net can fuse the null convolution to expand the sensory field, however, it still performs skip connection in a simple tandem manner and cannot fully utilize the original features. (4) The segmentation method using ResU-Net solves the problem of U-Net disconnection due to improving the simple tandem of skip connection into residual modules. The improved methods learn the target structure focusing on different shapes and sizes, and performing context extraction in coarser scales, thus solving the problem of U-Net disconnection, and the segmentation of distal vessels is also better than U-Net. However, over-segmentation occurs due to the redundancy of learning information phenomenon. (5) The segmentation method using DBCU-Net uses bidirectional convLSTM in the skip connection part, thus taking into account the forward and backward dependence of information, and the segmentation results of vessels are more refined, as shown in the yellow marked part of the figure, and the over-segmentation phenomenon is also greatly improved.\u003c/p\u003e \u003cp\u003eFinally, the segmentation results of the algorithm in this chapter are compared and analyzed with other segmentation methods. In this paper, we show through experiments that DBCU-Net improves the over-segmentation problem of existing methods and also has a great improvement in different segmentation evaluation indexes. As shown in Table.1, our method achieves optimal results on parameters such as Accuracy, Precision, Recall and F1 score compared with other methods. However, due to the imbalance of positive and negative samples in the dataset, this paper does not score the best on the AUC evaluation, which points out the robustness of the model proposed in this paper still needs to be improved.\u003c/p\u003e \u003cp\u003eThe large cost of labeling CAG images has added challenges to the acquisition of datasets. In the future, we hope to focus on unsupervised segmentation [\u003cspan citationid=\"CR25\" class=\"CitationRef\"\u003e25\u003c/span\u003e] to reduce the difficulty of dataset acquisition. For a single CAG image to segment blood vessels, we may use CAG dynamic video in the future [\u003cspan citationid=\"CR26\" class=\"CitationRef\"\u003e26\u003c/span\u003e] to segment blood vessels to reduce the effects of blood vessel crossing, masking, and noise on the segmentatio\u003c/p\u003e"},{"header":"6 Conclusion","content":"\u003cp\u003eWe proposed a new convolutional neural network architecture, which is based on U-Net and combines dense connectivity and bidirectional convLSTM. The new neural network is referred to as DBU-Net. Our model is proposed in this paper is a way to enhance the current approach over segmentation with insufficient accuracy. After the evaluation of the ablation experiment, the segmentation results of the blood vessels in this method are more accurate. We conduct our experiment on our private dataset, and achieve average Accuracy, Precision, Recall and F1-score for coronary artery segmentation of 0.985, 0.913, 0.847 and 0.879 respectively. In the mainstream metrics, our method has achieved competitive results.\u003c/p\u003e"},{"header":"Declarations","content":"\u003cp\u003e\u003cstrong\u003eCompeting Interests\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eWe declare that we have no financial and personal relationships with other people or organizations that can inappropriately influence our work, there is no professional or other personal interest of any nature or kind in any product, service and/or company that could be construed as influencing the position presented in, or the review of, the manuscript entitled.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eAuthor Contributions\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eYuqiang Shen: Resources, Data curation, Project administration. Zhe Chen: Methodology, Formal analysis, Writing - original draft. Jijun Tong: Conceptualization, Methodology. Nan Jiang: Writing - review \u0026amp; editing, Visualization. Yun Ning: Software, Validation.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eEthics approval\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThis is an observational study.\u0026nbsp;The data used in this article is obtained with the permission of the patients themselves.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eConsent to participate\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eInformed consent was obtained from all individual participants included in the study.\u0026rdquo;\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eConsent to publish\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThe authors affirm that human research participants provided informed consent for publication of the images in Figure(s) 1, 6 and 7.\u003c/p\u003e"},{"header":"References","content":"\u003col\u003e\u003cli\u003e\u003cspan\u003eSung H, Ferlay J, Siegel RL, Laversanne M, Soerjomataram I, Jemal A, Bray F (2021) Global cancer statistics 2020: GLOBOCAN estimates of incidence and mortality worldwide for 36 cancers in 185 countries. Cancer J Clin 71:209\u0026ndash;249. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.3322/caac.21660\u003c/span\u003e\u003cspan address=\"10.3322/caac.21660\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eMoayyedi PM, Lacy BE, Andrews CN, Enns RA, Howden CW, Vakil N (2017) ACG and CAG clinical guideline: management of dyspepsia. Official J Am Coll Gastroenterology| Acg 112:988\u0026ndash;1013. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1038/ajg.2017.154\u003c/span\u003e\u003cspan address=\"10.1038/ajg.2017.154\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eDehkordi MT, Sadri S, Doosthoseini A (2011) A review of coronary vessel segmentation algorithms. J Med Signals Sens 1:49. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.4103/2228-7477.83519\u003c/span\u003e\u003cspan address=\"10.4103/2228-7477.83519\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eSukanya A, Rajeswari R, Subramaniam Murugan K (2020) Region based coronary artery segmentation using modified Frangi's vesselness measure. Int J Imaging Syst Technol 30:716\u0026ndash;730. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1002/ima.22412\u003c/span\u003e\u003cspan address=\"10.1002/ima.22412\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eMabrouk S, Oueslati C, Ghorbel F (2017) Multiscale graph cuts based method for coronary artery segmentation in angiograms. Irbm 38:167\u0026ndash;175. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1016/j.irbm.2017.04.004\u003c/span\u003e\u003cspan address=\"10.1016/j.irbm.2017.04.004\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eDaoudi A, Mahmoudi S (2016) A fully automatic cardiac segmentation method using region growing technique[C]//Proceedings of the Mediterranean Conference on Pattern Recognition and Artificial Intelligence, : 103\u0026ndash;108\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eKerkeni A, Benabdallah A, Manzanera A, Bedoui MH (2016) A coronary artery segmentation method based on multiscale analysis and region growing. Comput Med Imaging Graph 48:49\u0026ndash;61. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1016/j.compmedimag.2015.12.004\u003c/span\u003e\u003cspan address=\"10.1016/j.compmedimag.2015.12.004\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eMa G, Yang J, Huang Y, Zhao H (2019) A novel automatic coronary artery segmentation method based on region growing with annular and spherical sector partition. J Med Imaging Health Inf 9:148\u0026ndash;152. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1166/jmihi.2019.2553\u003c/span\u003e\u003cspan address=\"10.1166/jmihi.2019.2553\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eHuang G, Liu Z, Van Der Maaten L, Weinberger KQ (2017) Densely connected convolutional networks[C]//Proceedings of the Ieee Conference on Computer Vision and Pattern Recognition, : 4700\u0026ndash;4708\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eZheng S, Ristovski K, Farahat A, Gupta C (2017) Long short-term memory network for remaining useful life estimation[C]//2017 Ieee International Conference on Prognostics and Health Management (icphm): Ieee, : 88\u0026ndash;95\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eCui Z, Ke R, Pu Z, Wang Y (2018) Deep bidirectional and unidirectional LSTM recurrent neural network for network-wide traffic speed prediction. Arxiv Preprint Arxiv. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.48550/arXiv.1801.02143\u003c/span\u003e\u003cspan address=\"10.48550/arXiv.1801.02143\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e. :1801.02143\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eCiresan D, Giusti A, Gambardella L, Schmidhuber J (2012) Deep neural networks segment neuronal membranes in electron microscopy images. Adv Neural Inf Process Syst 25. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://dl.acm.org/doi/10.5555/2999325.2999452\u003c/span\u003e\u003cspan address=\"https://dl.acm.doi/10.5555/2999325.2999452\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eRonneberger O, Fischer P, Brox T (2015) U-net: Convolutional networks for biomedical image segmentation[C]//International Conference on Medical Image Computing and Computer-assisted Intervention: Springer, Cham, : 234\u0026ndash;241\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eBlaiech AG, Mansour A, Kerkeni A, Bedoui MH, Ben Abdallah A (2019) Impact of enhancement for coronary artery segmentation based on deep learning neural network[C]//Iberian Conference on Pattern Recognition and Image Analysis: Springer, Cham, : 260\u0026ndash;272\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eShi X, Du T, Chen S, Zhang H, Guan C, Xu B (2020) UENet: a novel generative adversarial network for angiography image segmentation[C]//2020 42nd Annual International Conference of the Ieee Engineering in Medicine \u0026amp; Biology Society (embc): Ieee, : 1612\u0026ndash;1615\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eXian Z, Wang X, Yan S, Yang D, Chen J, Peng C (2020) Main coronary vessel segmentation using deep learning in smart medical. Mathematical Problems in Engineering 2020. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1155/2020/8858344\u003c/span\u003e\u003cspan address=\"10.1155/2020/8858344\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eFan J, Du C, Song S, Cong W, Hao A, Yang J (2019) Enhanced subtraction image guided convolutional neural network for coronary artery segmentation[C]//Chinese Conference on Image and Graphics Technologies: Springer, : 625\u0026ndash;632\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eLi R, Bian G, Zhou X, Xie X, Ni Z, Hou Z (2020) CAU-net: A Novel Convolutional Neural Network for Coronary Artery Segmentation in Digital Substraction Angiography[C]//International Conference on Neural Information Processing: Springer, Cham, : 185\u0026ndash;196\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eSamuel PM, Veeramalai T (2021) VSSC Net: vessel specific skip chain convolutional network for blood vessel segmentation. Comput Methods Programs Biomed 198:105769. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1016/j.cmpb.2020.105769\u003c/span\u003e\u003cspan address=\"10.1016/j.cmpb.2020.105769\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eZhou C, Dinh TV, Kong H, Yap J, Yeo KK, Lee HK, Liang K (2021) Automated deep learning analysis of angiography video sequences for coronary artery disease. Arxiv Preprint Arxiv. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.48550/arXiv.2101.12505\u003c/span\u003e\u003cspan address=\"10.48550/arXiv.2101.12505\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e. :2101.12505\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eYang S, Kweon J, Kim Y (2019) Major vessel segmentation on x-ray coronary angiography using deep networks with a novel penalty loss function[C]//International Conference on Medical Imaging with Deep Learning\u0026ndash;extended Abstract Track,\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eJun TJ, Kweon J, Kim Y, Kim D (2020) T-net: Nested encoder\u0026ndash;decoder architecture for the main vessel segmentation in coronary angiography. Neural Netw 128:216\u0026ndash;233. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1016/j.neunet.2020.05.002\u003c/span\u003e\u003cspan address=\"10.1016/j.neunet.2020.05.002\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eSong H, Wang W, Zhao S, Shen J, Lam K (2018) Pyramid dilated deeper convlstm for video salient object detection[C]//Proceedings of the European Conference on Computer Vision (eccv): Springer, Cham, : 715\u0026ndash;731\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eZhang Z (2018) Improved adam optimizer for deep neural networks[C]//2018 Ieee/acm 26th International Symposium on Quality of Service (iwqos): Springer International Publishing, : 1\u0026ndash;2\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eYu F, Zhao J, Gong Y, Wang Z, Li Y, Yang F, Dong B, Li Q, Zhang L (2019) Annotation-free cardiac vessel segmentation via knowledge transfer from retinal images[C]//International Conference on Medical Image Computing and Computer-assisted Intervention: Springer, : 714\u0026ndash;722\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eZhang J, Wang G, Xie H, Zhang S, Huang N, Zhang S, Gu L (2020) Weakly supervised vessel segmentation in X-ray angiograms by self-paced learning from noisy labels with suggestive annotation. Neurocomputing 417:114\u0026ndash;127. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1016/j.neuc\u003c/span\u003e\u003cspan address=\"10.1016/j.neuc\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003c/ol\u003e"}],"fulltextSource":"","fullText":"","funders":[],"hasAdminPriorityOnWorkflow":false,"hasManuscriptDocX":true,"hasOptedInToPreprint":true,"hasPassedJournalQc":"","hasAnyPriority":false,"hideJournal":false,"highlight":"","institution":"","isAcceptedByJournal":true,"isAuthorSuppliedPdf":false,"isDeskRejected":"","isHiddenFromSearch":false,"isInQc":false,"isInWorkflow":false,"isPdf":false,"isPdfUpToDate":true,"isWithdrawnOrRetracted":false,"journal":{"display":true,"email":"[email protected]","identity":"the-international-journal-of-cardiovascular-imaging","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":false,"externalIdentity":"caim","sideBox":"Learn more about [The International Journal of Cardiovascular Imaging](https://www.springer.com/journal/10554)","snPcode":"10554","submissionUrl":"https://submission.nature.com/new-submission/10554/3","title":"The International Journal of Cardiovascular Imaging","twitterHandle":"","acdcEnabled":true,"dfaEnabled":true,"editorialSystem":"em","reportingPortfolio":"Springer Hybrid","inReviewEnabled":true,"inReviewRevisionsEnabled":false},"keywords":"Coronary angiography, U-Net, Vessel segmentation, Deep Learning","lastPublishedDoi":"10.21203/rs.3.rs-2028133/v1","lastPublishedDoiUrl":"https://doi.org/10.21203/rs.3.rs-2028133/v1","license":{"name":"CC BY 4.0","url":"https://creativecommons.org/licenses/by/4.0/"},"manuscriptAbstract":"\u003cp\u003eCoronary angiography (CAG) is the \u0026ldquo;gold standard\u0026rdquo; for diagnosing coronary artery disease (CAD). However, due to the limitation of current imaging methods, the CAG image has low resolution and poor contrast with a lot of artifacts and noise, which makes it difficult for blood vessels segmentation. In this paper, we propose a DBCU-Net for automatic segmentation of CAG images, which is an extension of U-Net, DenseNet with bidirectional convLSTM. The main contribution of our network is that instead of convolution in the feature extraction of U-Net, we incorporate dense connectivity and the bidirectional convLSTM to highlight salient features. We conduct our experiment on our private dataset, and achieve average Accuracy, Precision, Recall and F1-score for coronary artery segmentation of 0.985, 0.913, 0.847 and 0.879 respectively.\u003c/p\u003e","manuscriptTitle":"DBCU-Net: Deep Learning Approach for Segmentation of Coronary Angiography Images","msid":"","msnumber":"","nonDraftVersions":[{"code":1,"date":"2022-09-07 14:45:31","doi":"10.21203/rs.3.rs-2028133/v1","editorialEvents":[{"type":"communityComments","content":0},{"type":"decision","content":"Major revision","date":"2023-02-14T07:56:36+00:00","index":"","fulltext":""},{"type":"editorInvitedReview","content":"","date":"2023-01-28T16:45:35+00:00","index":"hide","fulltext":""},{"type":"reviewerAgreed","content":"44e6dab6-a4b1-40d6-bed7-80d246ddf85d","date":"2023-01-03T06:55:58+00:00","index":"hide","fulltext":""},{"type":"reviewerAgreed","content":"b1142327-1175-4388-8d8d-57674807ac9e","date":"2022-12-12T08:06:08+00:00","index":"hide","fulltext":""},{"type":"reviewerAgreed","content":"5aed3021-9683-4397-8aba-234935a77534","date":"2022-10-02T10:28:58+00:00","index":"hide","fulltext":""},{"type":"editorInvitedReview","content":"","date":"2022-09-13T14:24:09+00:00","index":"hide","fulltext":""},{"type":"reviewerAgreed","content":"1bfddce4-4a56-452c-ad6f-7651ab1d24a3","date":"2022-09-05T13:46:35+00:00","index":"hide","fulltext":""},{"type":"reviewersInvited","content":"","date":"2022-09-03T16:13:48+00:00","index":"","fulltext":""},{"type":"editorAssigned","content":"","date":"2022-09-03T11:05:38+00:00","index":"","fulltext":""},{"type":"checksComplete","content":"","date":"2022-09-03T11:05:37+00:00","index":"","fulltext":""},{"type":"submitted","content":"The International Journal of Cardiovascular Imaging","date":"2022-09-03T09:06:11+00:00","index":"","fulltext":""}],"status":"published","journal":{"display":true,"email":"[email protected]","identity":"the-international-journal-of-cardiovascular-imaging","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":false,"externalIdentity":"caim","sideBox":"Learn more about [The International Journal of Cardiovascular Imaging](https://www.springer.com/journal/10554)","snPcode":"10554","submissionUrl":"https://submission.nature.com/new-submission/10554/3","title":"The International Journal of Cardiovascular Imaging","twitterHandle":"","acdcEnabled":true,"dfaEnabled":true,"editorialSystem":"em","reportingPortfolio":"Springer Hybrid","inReviewEnabled":true,"inReviewRevisionsEnabled":false}}],"origin":"","ownerIdentity":"9487fa6f-5f97-466d-9969-ad9297843cd6","owner":[],"postedDate":"September 7th, 2022","published":true,"recentEditorialEvents":[],"rejectedJournal":[],"revision":"","amendment":"","status":"published-in-journal","subjectAreas":[],"tags":[],"updatedAt":"2023-10-16T20:25:33+00:00","versionOfRecord":{"articleIdentity":"rs-2028133","link":"https://doi.org/10.1007/s10554-023-02849-3","journal":{"identity":"the-international-journal-of-cardiovascular-imaging","isVorOnly":false,"title":"The International Journal of Cardiovascular Imaging"},"publishedOn":"2023-04-05 20:22:43","publishedOnDateReadable":"April 5th, 2023"},"versionCreatedAt":"2022-09-07 14:45:31","video":"","vorDoi":"10.1007/s10554-023-02849-3","vorDoiUrl":"https://doi.org/10.1007/s10554-023-02849-3","workflowStages":[]},"version":"v1","identity":"rs-2028133","journalConfig":"researchsquare"},"__N_SSP":true},"page":"/article/[identity]/[[...version]]","query":{"redirect":"/article/rs-2028133","identity":"rs-2028133","version":["v1"]},"buildId":"WrCJVZZCHTDjtuVLN7oU0","isFallback":false,"isExperimentalCompile":false,"dynamicIds":[84888],"gssp":true,"scriptLoader":[]}