CVD are the leading cause of death globally according to WHO. About 17.9 million people died from CVDs in 2016, from CVD, mainly from heart disease and stroke111https://www.who.int/cardiovascular_diseases/about_cvd/en/. The number is still increasing annually. In recent decades, major advances have been made in cardiovascular research and practice aiming to improve diagnosis and treatment of cardiac diseases as well as reducing the mortality of CVD. Modern medical imaging techniques such as MRI, CT and US are now widely used, which enable non-invasive qualitative and quantitative assessment of cardiac anatomical structures and functions and provide support for diagnosis, disease monitoring, treatment planning and prognosis.
Of particular interest, cardiac image segmentation is an important first step in numerous applications. It partitions the image into a number of semantically (i.e. anatomically) meaningful regions, based on which quantitative measures can be extracted, such as the myocardial mass, wall thickness, LV and RV volume as well as EF etc. Typically, the anatomical structures of interest for cardiac image segmentation include the LV, RV, LA, RA, and coronary arteries. An overview of typical tasks related to cardiac image segmentation is presented in Fig. 1, where applications for the three most commonly used modalities, i.e., MRI, CT and US, are shown.
Before the rise of deep learning, traditional machine learning techniques such as model-based methods (e.g. active shape and appearance models) and atlas-based methods had been shown to achieve good performance in cardiac image segmentation(Petitjean_2015_MedIA; Peng_2016_MAGMA; Tavakoli_2013_CVIU; Lesage_2009_MedIA). However, they often require significant feature engineering or prior knowledge to achieve satisfactory accuracy. In contrast, DL-based algorithms are good at automatically
discovering intricate features from data for object detection and segmentation. These features are directly learned from data using a general-purpose learning procedure and in end-to-end fashion. This makes DL-based algorithms easy to apply to other image analysis applications. Benefiting from advanced computer hardware (e.g. graphical processing units (GPUs) and tensor processing units (TPUs)) as well as increased available data for training, DL-based segmentation algorithms have gradually outperformed previous state-of-the-art traditional methods, gaining more popularity in research. This trend can be observed in Fig.2A, which shows how the number of DL-based papers for cardiac image segmentation has increased strongly in the last years. In particular, the number of the publications for MR image segmentation is significantly higher than the numbers of the other two domains, especially in 2017. One reason, which can be observed in Fig. 2B, is that the publicly available data for MR segmentation has increased remarkably since 2016.
In this paper, we provide an overview of state-of-the-art deep learning techniques for cardiac image segmentation in the three most commonly used modalities (i.e. MRI, CT, US) in clinical practice and discuss the advantages and remaining limitations of current deep learning-based segmentation methods that hinder widespread clinical deployment. To our knowledge, there have been several review papers that presented overviews about applications of DL-based methods for general medical image analysis (Greenspan_2016_TMI; Shen_2017_Review; LITJENS_2017_MedIA), as well as some surveys dedicated to applications designed for cardiovascular image analysis (Gandhi_2018_Echocardiography; Mazurowski_2019_JMRI). However, none of them has provided a systematic overview focused on cardiac segmentation applications. This review paper aims at providing a comprehensive overview from the debut to the state-of-the-art of deep learning algorithms, focusing on a variety of cardiac image segmentation tasks (e.g. the LV, RV, and vessel segmentation) (Sec. 4). Particularly, we aim to cover most influential DL-related works in this field published until 1st August 2019 and categorized these publications in terms of specific methodology. Besides, in addition to the basics of deep learning introduced in Sec.3, we also provide a summary of public datasets (see Table 6) as well as public code (see Table 7), aiming to present a good reading basis for newcomers to the topic and encourage future contributions. More importantly, we provide insightful discussions about the current research situations (Sec.4.4) as well as challenges and potential directions for future work (Sec. 5).
Search criterion To identify related contributions, search engines like Scopus and PubMed were queried for papers containing (“convolutional” OR “deep learning”) and (“cardiac”) and (”image segmentation”) in title or abstract. Additionally, conference proceedings for MICCAI, ISBI and EMBC were searched based on the titles of papers. Papers which do not primarily focus on segmentation problems were excluded. The last update to the included papers was on Aug 1, 2019.
3 Fundamentals of Deep Learning
Deep learning models are deep artificial neural networks. Each neural network consists of an input layer, an output layer, and multiple hidden layers. In the following section, we will review several deep learning networks and key techniques that have been commonly used in state-of-the-art segmentation algorithms. For a more detailed and thorough illustration of the mathematical background and fundamentals of deep learning we refer the interested reader to Goodfellow_2016_MIT.
3.1 Neural Networks
In this section, we first introduce basic neural network architectures and then briefly introduce building blocks which are commonly used to boost the ability of the networks to learn features that are useful for image segmentation.
3.1.1 Convolutional Neural Networks (CNNs)
In this part, we will introduce CNN, which is the most common type of deep neural networks for image analysis. CNN have been successfully applied to advance the state-of-the-art on many image classification, object detection and segmentation tasks.
As shown in Fig. 3A, a standard CNN consists of an input layer, an output layer and a stack of functional layers in between that transform an input into an output in a specific form (e.g. vectors). These functional layers often contains convolutional layers, pooling layers and/or fully-connected layers. In general, each convolution uses a kernel (for 2D input) or
kernel (for 3D input) followed by batch normalization(Ioffe_2015_ICML)
after which the output is passed through a nonlinear activation function (e.g. ReLU), which is used to extract feature maps from an image. These feature maps are then downsampled by pooling layers, typically by a factor of 2, which removes redundant features to improve the statistical efficiency and model generalization. After that, fully connected layers are applied to reduce the dimension of features and find the most task-relevant features for inference. The output of the network is a fix-sized vector where each element can be a probabilistic score for each category (for image classification), a real value for a regression task (e.g. the left ventricular volume estimation) or a set of values (e.g. the coordinates of a bounding box for object detection and localization).
In general, the size of convolution kernel is chosen to be small in general, e.g.
, in order to reduce computational costs. While the kernels are small, one can increase the receptive field (the area of the input image that potentially impacts the activation of a particular convolutional kernel/neuron) by increasing the number of convolutional layers. For example, a convolutonal layer with largekernels can be replaced by three layers with small kernels. The number of parameters is reduced by a factor of while the receptive field remains the same (). An online resource 222https://fomoro.com/research/article/receptive-field-calculator is referred here, which illustrates and visualizes the change of receptive field by varying the number of hidden layers and the size of kernels. In general, increasing the depth of convolution neural networks (the number of hidden layers) to enlarge the receptive field can lead to improved model performance, e.g. classification accuracy (Simonyan_2015_ICLR).
CNNs for image classification can also be employed for image segmentation applications without major adaptations to the network architecture (Ciresan_2012_NIPS), as shown in Fig. 3B. However, this requires to divide each image into patches and then train a CNN to predict the class label of the center pixel for every patch. One major disadvantage of this patch-based approach is that, at inference time, the network has to be deployed for every patch individually despite the fact that there is a lot of redundancy due to multiple overlapping patches in the image. As a result of this inefficiency, the main application of CNNs with fully connected layers is object localization, which aims to estimate the bounding box of the object of interest in an image. This bounding box is then used to crop the image, forming an image pre-processing step to reduce the computational cost for segmentation (Avendi_2016_MedIA). For efficient, end-to-end pixel-wise segmentation, a variant of CNNs called FCN is more commonly used, which will be discussed in the next section.
3.1.2 Fully Convolutional Neural Networks (FCNs)
The idea of FCN was first introduced by Long_2014_CVPR for image segmentation. FCNs are a special type of CNN that do not have any fully connected layers. In general, as shown in Fig. 4A, FCN are designed to have an encoder-decoder structure such that they can take input of arbitrary size and produce the output with the same size. Given an input image, the encoder first transforms the input into high-level feature representation whereas the decoder interprets the feature maps and recovers spatial details back to the image space for pixel-wise prediction through a series of transposed convolution and convolution operations. Here, transposed convolutions are used for up-scaling the feature maps, typically by a factor of 2. These transposed convolutions can also be replaced by unpooling layers and upsampling layers. Compared to a patch-based CNN for segmentation, FCN is trained and applied to the entire images, removing the need for patch selection (Shelhamer_2017_TPAMI).
FCN with the simple encoder-decoder structure in Fig. 4A may be limited to capture detailed context information in an image for precise segmentation as some features may be eliminated by the pooling layers in the encoder. Several variants of FCN have been proposed to propagate features from the encoder to the decoder, in order to boost the segmentation accuracy. The most well-known and most popular variant of FCN for biomedical image segmentation is the U-net (Ronneberger_2015_MICCAI). On the basis of the vanilla FCN (Long_2014_CVPR), the U-net employs skip connections between the encoder and decoder to recover spatial context loss in the down-sampling path, yielding more precise segmentation (see Fig. 4B). Several state-of-the-art cardiac image segmentation methods have adopted the U-net or its 3D variants, the 3D U-net (Cicek_2016_MICCAI) and the 3D V-net (Milletari_2016_3DV), as their backbone networks, achieving promising segmentation accuracy for a number of cardiac segmentation tasks (Tao_2019_Radiology; Isensee_2017_STACOM; Xia_2018_STACOM).
3.1.3 Recurrent Neural Networks (RNNs)
RNN are another type of artificial neural networks which are used for sequential data, such as cine MRI and ultrasound image sequences. An RNN can ‘remember’ the past and use the knowledge learned from the past to make its present decision, see Fig 5. For example, given a sequence of images, an RNN takes the first image as input, captures the information to make a prediction and then memorize this information which is then utilized to make a prediction for the next image. The two most widely used architectures in the family of RNN are LSTM (Hochreiter_1997_NC) and GRU (Cho_2014_EMNLP), which are capable of modeling long-term memory. A use case for cardiac segmentation is to combine an RNN with a 2D FCN so that the combined network is capable of capturing information from adjacent slices to improve the inter-slice coherence of segmentation results (Poudel_2016_HVSCMR).
3.1.4 Autoencoders (AE)
AE are a type of neural networks that are designed to learn compact latent representations from data without supervision. A typical architecture of an autoencoder consists of two networks: an encoder network and a decoder network for the reconstruction of the input, see Fig.6. Since the learned representations contain generally useful information in the original data, many researchers have employed autoencoders to extract general semantic features or shape information from input images or labels and then use those features to guide the cardiac image segmentation (Oktay_2016_MICCAI; Schlemper_2018_MICCAI; Yue_2019_MICCAI).
3.1.5 Generative Adversarial Networks (GAN)
The concept of GAN was proposed by Goodfellow_2014_NIPS for image synthesis from noise. GANs are a type of generative models that learn to model the data distribution of real data and thus are able to create new image examples. As shown in Fig. 7A, a GAN consists of two networks: a generator network and a discriminator network. During training, the two networks are trained to compete against each other: the generator produces fake images aimed at fooling the discriminator, whereas the discriminator tries to identify real images from fake ones. This type of training is referred to as ‘adversarial training’, since the two models are both set to win the competition. This training scheme can also be used for training a segmentation network. As shown in Fig. 7B, the generator is replaced by a segmentation network and the discriminator is required to distinguish the generated segmentation maps from the ground truth ones (the target segmentation maps). In this way, the segmentation network is encouraged to produce more anatomically plausible segmentation maps (Luc_2016_NIPS_workshop; Savioli_2018_Arxiv).
3.1.6 Advanced building blocks for improved segmentation
Medical image segmentation, as an important step for quantitative analysis and clinical research, requires a pixel-wise accuracy. Over the past years, many researchers have developed advanced building blocks to learn robust, representative features for precise segmentation. These techniques have been widely applied to state-of-the-art neural networks (e.g. U-net) to improve cardiac image segmentation performance. Therefore, we identified several important techniques reported in the literature to this end and present them with corresponding references for further reading. These techniques are:
Advanced convolutional modules for multi-scale feature aggregation in the hidden layers:
Inception modules (Szegedy_2015_CVPR), see Fig. 8A;
Dilated convolutional kernels (Yu_2015_ICLR);
Deep supervision (Lee_2015_AIS);
Atrous spatial pyramid pooling (Chen_2017_Arxiv);
Adaptive convolutional kernels designed to pay attention to important features:
Attention units (Vaswani_2017_NIPS), see Fig. 8B;
Squeeze-and-excitation blocks (Hu_2018_CVPR);
Interlayer connections designed to reuse features from previous layers:
Residual connections (He_2016_CVPR), see Fig. 8C;
Dense connections (Huang_2017_CVPR).
3.2 Training Neural Networks
Before being able to perform inference, neural networks must be trained. This training process requires a dataset that contains paired images and labels
for training and testing, an optimizer (e.g. stochastic gradient descent, Adam) and a loss function to update the model parameters. This function accounts for the error of the network prediction in each iteration during training, providing signals for the optimizer to update the network parameters through backpropagation. The goal of training is to find proper values of the network parameters to minimize the loss function.
3.2.1 Common Loss Functions
For regression tasks (e.g. heart localization, calcium scoring, landmark detection, image reconstruction), the simplest loss function is the MSE:
where is the vector of target values and is the vector of the predicted values; is the number of data samples.
Cross-entropy is the most common loss for both image classification and segmentation tasks. In particular, the cross-entropy loss for segmentation summarizes the pixel-wise probability errors between the predicted probabilistic output and its corresponding target segmentation map for each class :
where is the number of all classes. Another loss function which is specifically designed for object segmentation is called soft-Dice loss function (Milletari_2016_3DV), which penalizes the mismatch between a predicted segmentation map and its target map at pixel-level:
In addition, there are several variants of the cross-entropy or soft-Dice loss such as the weighted cross-entropy loss (Jang_2017_STACOM; Baumgartner_2017_STACOM) and weighted soft-Dice loss (Yang_2017_STACOM; Khened_2019_MedIA) that are used to address potential class imbalance problem in medical image segmentation tasks where the loss term is weighted to account for rare classes or small objects.
3.2.2 Reduce over-fitting
The biggest challenge of training deep networks for medical image analysis is over-fitting, due to the fact that there is often a limited number of training images in comparison with the number of learnable parameters in a deep network. A number of techniques have been developed to alleviate this problem. Some of the techniques are the following ones:
Weight initialization (He_2015_ICCV) and weight regularization (i.e. L1/L2 regularization)
Ensemble learning (Kamnitsas_2017_Arxiv)
Data augmentation by artificially generating training samples via affine transformations
Transfer learning with a model pre-trained on existing large datasets.
3.3 Evaluation Metrics
To quantitatively evaluate the performance of automated segmentation algorithms, three types of metrics are commonly used: a) volume-based metrics (e.g. Dice metric, Jaccard similarity index); b) surface distance-based metrics (e.g. mean contour distance, Hausdorff distance); c) clinical performance metrics (e.g. ventricular volume and mass). For a detailed illustration of common used clinical indices in cardiac image analysis, we recommend the review paper by Peng_2016_MAGMA. In our paper, we mainly report the accuracy of methods in terms of the Dice metric for ease of comparison. The Dice score measures the ratio of overlap between two results (e.g. automatic segmentation vs manual segmentation), ranging from 0 (mismatch) to 1 (perfect match).
4 Deep Learning for Cardiac Image Segmentation
|Application||Selected works||Description||Type of Images||Structure(s)|
|Lieman-Sifry_2017_FIMH||A lightweight FCN (E-Net)||SAX||Bi-ventricle|
|Isensee_2017_STACOM||2D U-net +3D U-net (ensemble)||SAX||Bi-ventricle|
|Jang_2017_STACOM||2D M-Net with weighted cross entropy loss||SAX||Bi-ventricle|
|Baumgartner_2017_STACOM||2D U-net with cross entropy||SAX||Bi-ventricle|
|SAX, 2CH, 4CH||Four chambers|
|Khened_2019_MedIA||2D Dense U-net with inception module||SAX||Bi-ventricle|
|Introducing spatial or temporal context|
|Poudel_2016_HVSCMR||2D FCN with RNN to model inter-slice coherency||SAX||Bi-ventricle|
|Patravali_2017_STACOM||2D multi-channel FCN to aggregate inter-slice information||SAX||Bi-ventricle|
|Wolterink_2017_STACOM||Dilated U-net to segment ED and ES simultaneously||SAX||Bi-ventricle|
|Applying anatomical constraints|
|Oktay_2018_TMI||FCN trained with additional anatomical shape-based regularization||SAX;US||LV|
|SAX, 2CH, 4CH||Four chambers|
|Hybrid segmentation methods|
|Yang_2016_MICCAI||CNN combined with Multi-atlas||SAX||LV|
Level-set based segmentation with Deep belief networks
|Atrial Segmentation||Mortazi_2017_STACOM||Multi-view CNN with adaptive fusion strategy||3D scans||LA|
|Xiong_2019_TMI||Patch-based dual-stream 2D FCN||LGE MRI||LA|
|Xia_2018_STACOM||Two-stage pipeline; 3D U-net (localization) +3D U-net (segmentation)||LGE MRI||LA|
|LGE MRI||LA; atrial scars|
|LGE MRI||LA; atrial scars|
|LGE MRI||Myocardial scars|
|LGE MRI||Myocardial scars|
|cine MRI||Myocardial scars|
|Whole Heart Segmentation||Yu_2017_MICCAI||3D U-net with deep supervision||3D scans||Blood pool+Myocardium of the heart|
|Li_2017_RSAM||3D FCN with deep supervision||3D scans||Blood pool+Myocardium of the heart|
|Wolterink_2017_RSAM||dilated CNN with deep supervision||3D scans||Blood pool+Myocardium of the heart|
In this section, we provide a summary of deep learning-based applications for the three main imaging modalities: MRI, CT, and US regarding specific applications for targeted structures. In general, these deep learning-based methods provide an efficient and effective way to segmenting particular organs or tissues (e.g. the LV, coronary vessels, scars) in different modalities, facilitating follow-up quantitative analysis of cardiovascular structure and function. Among these works, a large portion of these methods are designed for ventricle segmentation, especially in MR and US domains. The objective of ventricle segmentation is to delineate the endocardium and epicardium of the LV and/or RV. These segmentation maps are important for deriving clinical indices, such as LVEDV, LVESV, RVEDV, RVESV, and EF. In addition, these segmentation maps are essential for 3D shape analysis (Xue_2018_MedIA; Biffi_2018_MICCAI), 3D+time motion analysis (Zheng_2019_MedIA) and survival prediction (Bello_2019_NMI).
4.1 Cardiac MR Image Segmentation
Cardiac MRI is a non-invasive imaging technique that can visualize the structures within and around the heart. Compared to CT, it does not require ionising radiation. Instead, it relies on the magnetic field in conjunction with radio-frequency waves to excite hydrogen nuclei in the heart, and then generates an image by measuring their response. By utilizing different imaging sequences, cardiac MRI allows accurate quantification of both cardiac anatomy and function (e.g. cine imaging) and pathological tissues such as scars (LGE imaging). Accordingly, cardiac MRI is currently regarded as the gold standard for quantitative cardiac analysis (Van_der_Geest_1999_JMRI).
A group of representative deep learning based cardiac MR segmentation methods are shown in Table 1. From the table, one can see that a majority of works have focused on segmenting cardiac chambers (e.g. LV, RV, LA). In contrast, there are relatively fewer works on segmenting abnormal cardiac tissue regions such as myocardial scars and atrial fibrosis. This is likely due to the limited relevant public datasets as well as the difficulty of the task. In addition, to the best of our knowledge, there are very few works that apply deep learning techniques to atrial wall segmentation, as also suggested by a recent survey paper (Karim_2018_MedIA). In the following sections, we will describe and discuss these methods regarding different applications in detail.
4.1.1 Ventricle Segmentation
Vanilla FCN-based Segmentation: Tran_2016_Arxiv was among the first ones to apply a FCN (Shelhamer_2017_TPAMI) to segment the left ventricle, myocardium and right ventricle directly on short-axis cardiac MR images. Their end-to-end approach based on FCN achieved competitive segmentation performance, significantly outperforming traditional methods in terms of both speed and accuracy. In the following years, a number of works based on FCN have been proposed, aiming at achieving further improvements in segmentation performance. In this regard, one stream of work focuses on optimizing the network structure to enhance the feature learning capacity for segmentation (Khened_2019_MedIA; Li_2019_ITBE; Zhou_2018_RAL; Zhang_2019_Access; Cong_2018_JE; Jang_2017_STACOM; Fahmy_2019_JCMR). For example, Khened_2019_MedIA developed a dense U-net with inception modules to combine multi-scale features for robust segmentation across images with large anatomical variability. Jang_2017_STACOM; Yang_2017_STACOM; Sander_2019_MIP; Chen_2019_ISBI investigated different loss functions such as weighted cross-entropy, weighted Dice loss, deep supervision loss and focal loss to improve the segmentation performance. Among these FCN-based methods, the majority of approaches use 2D networks rather than 3D networks for segmentation. This is mainly due to the typical low through-plane resolution and motion artifacts of most cardiac MR scans, which limits the applicability of 3D networks (Baumgartner_2017_STACOM).
Introducing spatial or temporal context: One drawback of using 2D networks for cardiac segmentation is that these networks work slice by slice, and thus they do not leverage any inter-slice dependencies. As a result, 2D networks can fail to locate and segment the heart on challenging slices such as apical and basal slices where the contours of the ventricles are not well defined. To address this problem, a number of works have attempted to introduce additional contextual information to guide 2D FCN. This contextual information can include shape priors learned from labels or multi-view images (Zotti_2017_STACOM; Zotti_2019_JBHI; Chen_2019_MICCAI). Others extract spatial information from adjacent slices to assist the segmentation, using recurrent units (RNN) or multi-slice networks (2.5D networks) (Poudel_2016_HVSCMR; Patravali_2017_STACOM; Du_2019_JTEHM; Zheng_2018_TMI). These networks can also be applied to leveraging information across different temporal frames in the cardiac cycle to improve spatial and temporal consistency of segmentation results (Yan_2018_MICCAI; Savioli_2018_SNAMS; Du_2019_JTEHM; Qin_2018_MICCAI; Wolterink_2017_STACOM).
Applying anatomical constraints: Another problem that may limit the segmentation performance of both 2D and 3D FCN is that they are typically trained with pixel-wise loss functions only (e.g. cross-entropy or soft-Dice losses). These pixel-wise loss functions may not be sufficient to learn features that represent the underlying anatomical structures. Several approaches therefore focus on designing and applying anatomical constraints to train the network to improve its prediction accuracy and robustness. These constraints are represented as regularization terms which take into account the topology (Clough_2019_IPMI), contour and region information (Chen_2019_CVPR) or shape information (Oktay_2018_TMI; Yue_2019_MICCAI), encouraging the network to generate more anatomically plausible segmentations. In addition to regularizing networks at training time, Painchaud_2019_MICCAI proposed a variational AE to correct inaccurate segmentations, in the post-processing stage.
Multi-task learning: Multi-task learning has also been explored to regularize FCN-based cardiac ventricle segmentation during training by performing auxiliary tasks that are relevant to the main segmentation task, such as motion estimation (Qin_2018_MLMLR), estimation of cardiac function (Dangi_2018_STACOM), ventricle size classification (Zhang_2018_ICASSP) and image reconstruction (ChartsiasA_2018; Huang_2019_FIMH). Training a network for multiple tasks simultaneously encourages the network to extract features which are useful across these tasks, resulting in improved learning efficiency and prediction accuracy.
Multi-stage networks: Recently, there is a growing interest in applying neural networks in a multi-stage pipeline which breaks down the segmentation problem into subtasks (Vigneault_2018_MedIA; Zheng_2018_TMI; Li_2019_ISBI; Tan_2017_MedIA; Liao_2019_TCyber). For example, Zheng_2018_TMI; Li_2019_ISBI proposed a ROI localization network followed by a segmentation network. Likewise, Vigneault_2018_MedIA proposed a network called Omega-Net which consists of a U-net for cardiac chamber localization, a learnable transformation module to normalize image orientation and a series of U-nets for fine-grained segmentation. By explicitly localizing the ROI and by rotating the input image into a canonical orientation, the proposed method better generalizes to images with varying sizes and orientations.
Hybrid segmentation methods: Another stream of work aims at combining neural networks with classical segmentation approaches, e.g. level-sets (Ngo_2016_MedIA; Duan_2018_MICCAI), deformable models (Avendi_2016_MedIA; Avendi_2017_MRM; Medley_2019_ISBI), atlas-based methods (Yang_2016_MICCAI; Rohe_2017_STACOM) and graph-cut based methods (Lu_2019_ICCSP)
. Here, neural networks are applied in the feature extraction and model initialization stages, reducing the dependency on manual interactions and improving the segmentation accuracy of the conventional segmentation methods deployed afterwards. For example,Avendi_2016_MedIA proposed one of the first DL-based methods for LV segmentation in cardiac short-axis MR images. The authors first applied a CNN to automatically detect the LV and then used an AE to estimate the shape of the LV. The estimated shape was then used to initialize follow-up deformable models for shape refinement. As a result, the proposed integrated deformable model converges faster than conventional deformable models and the segmentation achieves higher accuracy. In their later work, the authors extended this approach to segment RV (Avendi_2017_MRM). While these hybrid methods demonstrated better segmentation accuracy than previous non-deep learning methods, most of them still require an iterative optimization for shape refinement. Furthermore, these methods are often designed for one particular anatomical structure. As noted in the recent benchmark study (Bernard_2018_TMI), most state-of-the-art segmentation algorithms for bi-ventricle segmentation are based on end-to-end FCN, which allows the simultaneous segmentation of the LV and RV.
To better illustrate these developments for cardiac ventricle segmentation from cardiac MR images, we collate a list of bi-ventricle segmentation methods that have been trained and tested on the ACDC dataset, reported in Table 2. For ease of comparison, we only consider those methods which have been evaluated on the same online test set (50 subjects). As the ACDC challenge organizers keep the online evaluation platform open to the public, our comparison not only includes the methods from the original challenge participants (summarized in the benchmark study paper from Bernard_2018_TMI) but also three segmentation algorithms that have been proposed after the challenge (i.e. Zotti_2019_JBHI; Li_2019_ISBI; Painchaud_2019_MICCAI). From this comparison, one can see that top algorithms are the ensemble method proposed by Isensee_2017_STACOM and the two-stage method proposed by Li_2019_ISBI, both of which are based on FCN. In particular, compared to the traditional level-set method (Tziritas_2017_STACOM), both methods achieved considerably higher accuracy even for the more challenging segmentation of the left ventricular myocardium (Myo), indicating the power of deep learning based approaches.
|Isensee_2017_STACOM||2D U-net +3D U-net (ensemble)||0.950||0.911||0.923|
|Li_2019_ISBI||Two 2D FCNs for ROI detection and segmentation respectively;||0.944||0.911||0.926|
|Zotti_2019_JBHI||2D GridNet-MD with registered shape prior||0.938||0.894||0.910|
|Khened_2019_MedIA||2D Dense U-net with inception module||0.941||0.894||0.907|
|Baumgartner_2017_STACOM||2D U-net with cross entropy loss||0.937||0.897||0.908|
|Zotti_2017_STACOM||2D GridNet with registered shape prior||0.931||0.890||0.912|
|Jang_2017_STACOM||2D M-Net with weighted cross entropy loss||0.940||0.885||0.907|
|Painchaud_2019_MICCAI||FCN followed by an AE for shape correction||0.936||0.889||0.909|
|Wolterink_2017_STACOM||Multi-input 2D dilated FCN, segmenting paired ED and ES frames simultaneously||0.940||0.885||0.900|
|Patravali_2017_STACOM||2D U-net with a Dice loss||0.920||0.890||0.865|
|Rohe_2017_STACOM||Multi-atlas based method combined with 3D CNN for registration||0.929||0.868||0.881|
|Tziritas_2017_STACOM||Level-set +MRF; Non-deep learning method||0.907||0.798||0.803|
|Yang_2017_STACOM||3D FCN with deep supervision||0.820||N/A||0.780|
Note that for simplicity, we report the average Dice scores for each structure over ED and ES phases. More detailed comparison for different phases can be found on the public leaderboard in the post testing part (https://acdc.creatis.insa-lyon.fr) as well as corresponding published works in this table.
4.1.2 Atrial Segmentation
AF is one of the most common cardiac electrical disorders, affecting around 1 million people in the UK 333https://www.nhs.uk/conditions/atrial-fibrillation/. Accordingly, atrial segmentation is of prime importance in the clinic, improving the assessment of the atrial anatomy in both pre-operative AF ablation planning and post-operative follow-up evaluations. In addition, the segmentation of atrium can be used as a basis for scar segmentation and atrial fibrosis quantification from LGE images. Traditional methods such as region growing (Karim_2008_Online) and methods that employ strong priors (i.e. atlas-based label fusion (Tao_2016_JMRI) and non-rigid registration (Zhuang_2010_TMI)) have been applied in the past for automated left atrium segmentation. However, the accuracy of these methods highly relies on good initialization and ad-hoc pre-processing methods, which limits the widespread adoption in the clinic.
Recently, Bai_2018_JCMR and Vigneault_2018_MedIA applied 2D FCN to directly segment the LA and RA from standard 2D long-axis images, i.e. 2CH, 4CH views. Notably, their networks can also be trained to segment ventricles from 2D short-axis stacks without any modifications to the network architecture. Likewise, Xiong_2019_TMI; Preetha_2018_STACOM; Bian_2018_STACOM; Chen_2018_STACOM applied 2D FCNs to segment the atrium from 3D LGE images in a slice-by-slice fashion, where they optimized the network structure for enhanced feature learning. 3D networks (Xia_2018_STACOM; Savioli_2018_STACOM; Jia_2018_STACOM; Vesal_2018_STACOM; Li_2018_STACOM) and multi-view FCN (Mortazi_2017_STACOM; Yang_2018_EMBC) have also been explored to capture 3D global information from 3D LGE images for accurate atrium segmentation.
In particular, Xia_2018_STACOM proposed a fully automatic two-stage segmentation framework which contains a first 3D U-net to roughly locate the atrial center from down-sampled images followed by a second 3D U-net to accurately segment the atrium in the cropped portions of the original images at full resolution. Their multi-stage approach is both memory-efficient and accurate, ranking first in the LASC 2018 with a mean Dice score of 0.93 evaluated on a test set of 54 cases.
4.1.3 Scar Segmentation
Scar characterization is usually performed using LGE MR imaging, a contrast-enhanced MR imaging technique. LGE MR imaging enables the identification of myocardial scars and atrial fibrosis, allowing improved management of myocardial infarction and atrial fibrillation (Kim_1999_Circulation). Prior to the advent of deep learning, scar segmentation was often performed using intensity thresholding-based or clustering methods which are sensitive to the local intensity changes (Zabihollahy_2018_MI). The main limitation of these methods is that they usually require the manual segmentation of the region of interest to reduce the search space and the computational costs (Carminati_2016_JTI). As a result, these semi-automated methods are not suitable for large-scale studies or clinical deployment.
Deep learning approaches have been combined with traditional segmentation methods for the purpose of scar segmentation: Yang_2017_MIUA; Yang_2018_MedPhy applied an atlas-based method to identify the left atrium and then applied deep neural networks to detect fibrotic tissue in that region. Relatively to end-to-end approaches, Chen_2018_MICCAI applied deep neural networks to segment both the left atrium and the atrial scars. In particular, the authors employed a multi-view CNN with a recursive attention module to fuse features from complementary views for better segmentation accuracy. Their approach achieved a mean Dice score of 0.90 for the LA region and a mean Dice score of 0.78 for atrial scars.
In the work of Fahmy_2018_JACC, the authors applied a U-net based network to segment the myocardium and the scars at the same time from LGE images acquired from patients with HCM, achieving a fast segmentation speed. However, the reported segmentation accuracy for the scar regions was relatively low (mean Dice: 0.58). Zabihollahy_2018_MI; Moccia_2019_Magma instead adopted a semi-automated method which requires a manual segmentation of the myocardium followed by the application of a 2D network to differentiate scars from normal myocardium. They reported higher segmentation accuracy on their test sets (mean Dice
0.68). At the moment, fully-automated scar segmentation is still a challenging task since the infarcted regions in patients can lead to kinematic variabilities and abnormalities in those contrast-enhanced images. Interestingly,Xu_2018_MedIA developed an RNN which leverages motion patterns to automatically delineate myocardial infarction area from cine MR image sequences without contrast agents. Their method achieved a high overall Dice score of 0.90 when compared to the manual annotations on LGE MR images, providing a novel approach for infarction assessment.
4.1.4 Aorta Segmentation
The segmentation of the aortic lumen from cine MR images is essential for accurate mechanical and hemodynamic characterization of the aorta. One common challenge for this task is the typical sparsity of the annotations in aortic cine image sequences, where only a few frames have been annotated. To address the problem, Bai_2018_MICCAI applied a non-rigid image registration method (Rueckert_1999_TMI) to propagate the labels from the annotated frames to the unlabeled neighboring ones in the cardiac cycle, effectively generating pseudo annotated frames that could be utilized for further training. This semi-supervised method achieved an average Dice metric of 0.96 for the ascending aorta and 0.95 for the descending aorta over a test set of 100 subjects. In addition, compared to a previous approach based on deformable models (Herment_2010_JMRI), their approach based on FCN and RNN can directly perform the segmentation task on a whole image sequence without requiring the explicit estimation of the ROI.
4.1.5 Whole Heart Segmentation
Apart from the above mentioned segmentation applications which target one particular structure, deep learning can also be used to segment the main substructures of the heart in 3D MR images (Yu_2017_MICCAI; Wolterink_2017_RSAM; Li_2017_RSAM; Shi_2018_MICCAI). An early work from Yu_2017_MICCAI adopted a 3D dense FCN to segment the myocardium and blood pool in the heart from 3D MR scans. Recently, more and more methods began to apply deep learning pipelines to segment more specific substructures (incl. four chambers, MYO, aorta, PV) in both 3D CT and MR images. This has been facilitated by the availability of public datasets for whole heart segmentation (MM-WHS). In general, the segmentation task on MR images is harder than the one of CT images mainly because of the large variations in terms of image intensity distribution among different scanners. As mentioned in a recent benchmark study paper by Zhuang_2019_MedIA, deep learning methods in general achieve better segmentation accuracy on CT images compared to that of MR images. We will discuss these segmentation methods in the next CT section in further detail (see section 4.2.1).
4.2 Cardiac CT Image Segmentation
CT is a non-invasive imaging technique that is performed routinely for disease diagnosis and treatment planning. In particular, cardiac CT scans are used for assessment of cardiac anatomy and specifically the coronary arteries. There are two main imaging modalities: non-contrast CT imaging and contrast-enhanced coronary CT angiography (CTA). Typically, non-contrast CT imaging exploits density of tissues to generate an image, such that different densities using various attenuation values such as soft tissues, calcium, fat, and air can be easily distinguished, and thus allows to estimate the amount of calcium present in the coronary arteries (Kang_2012_JoEI). In comparison, contrast-enhanced coronary CTA, which is acquired after the injection of a contrast agent, can provide excellent visualization of cardiac chambers, vessels and coronaries, and has been shown to be effective in detecting non-calcified coronary plaques. In the following sections, we will review some of the most commonly used deep learning-based cardiac CT segmentation methods. A summary of these approaches is presented in Table 3.
|Application||Selected works||Description||Imaging Modality||Structure(s)|
|Zreik_2016_ISBI_CT||patch based CNN||CTA||LV|
|Payer_2018_MMWHS||a pipeline of two FCNs||MR/CT||WHS|
|Tong_2017_MMWHS||deeply supervised 3D U-net||MR/CT||WHS|
|Wang_2018_arXiv||two-stage 3D U-net with dynamic ROI extraction||MR/CT||WHS|
|Xu_2018_arXiv||faster RCNN and U-net||CT||WHS|
|Wang_2017_MMWHS||orthogonal 2D U-nets with shape context||MR/CT||WHS|
|Mortazi_2017_MMWHS||multi-planar FCNs with an adaptive fusion strategy||MR/CT||WHS|
|Yang_2017c_MMWHS||3D U-net with deep supervision||MR/CT||WHS|
|Ye_2019_Access||3D deeply-supervised U-net with multi-depth fusion||CT||WHS|
|Joyce_2018_MIDL||unsupervised segmentation with GANs||MR/CT||LV/RV/Myo|
|Merkow_2016_MICCAI||3D U-net with deep multi-scale supervision||CTA||Vessel|
template transformer network
|CNN as pre-/post-processing|
|Gulsun_2016_MICCAI||CNN as path pruning||CTA||coronary artery centerline|
|Guo_2019_IPMI||multi-task FCN with a minimal patch extractor||CTA||coronary artery centerline|
|Shen_2019_IEEEAccess||3D FCN with level set||CTA||Vessel|
|Wolterink_2019_MedIA||CNN to estimate direction classification and radius regression||CTA||coronary artery centerline|
|Wolterink_2019_arXiv||graph convolutional network||CTA||Vessel|
|Lessmann_2017_TMI||two consecutive CNNs||CT||CAC|
|Liu_2018_STACOM||3D vessel-focused ConvNets||CTA||CAC/NCP/MCP|
|Santini_2017_IFMBE||patch based CNN||CT||CAC|
|Shadmi_2018_ISBI||U-net and FC DenseNet||CT||CAC|
4.2.1 Cardiac Substructure Segmentation
Accurate delineation of cardiac substructures plays a crucial role in cardiac function analysis, providing important clinical variables such as EF, myocardial mass, wall thickness etc. Typically, the cardiac substructures that are segmented include the LV, RV, LA, RA, MYO, AO and PA.
Two-step segmentation: One group of deep learning methods relies on a two-step segmentation procedure, where a ROI is first extracted and then fed into a CNN for subsequent classification (Zreik_2016_ISBI_CT; Dormer_2018_SPIE_CT). For instance, Zreik_2016_ISBI_CT proposed a two-step LV segmentation process where a bounding box for the LV is first detected using the method described in (De_2017_TMI), followed by a voxel classification within the defined bounding box using a patch-based CNN. More recently, FCN, especially U-net (Ronneberger_2015_MICCAI), has become the method of choice for cardiac CT segmentation. Zhuang_2019_MedIA provides a comparison of a group of methods (Payer_2018_MMWHS; Wang_2017_MMWHS; Yang_2017b_MMWHS; Yang_2017c_MMWHS; Tong_2017_MMWHS; Mortazi_2017_MMWHS) for WHS that have been evaluated on the MM-WHS challenge. Several of these methods (Payer_2018_MMWHS; Tong_2017_MMWHS; Xu_2018_arXiv; Wang_2018_arXiv) combine a localization network, which produces a coarse detection of the heart, with 3D FCNs applied to the detected ROI for segmentation. This allows the segmentation network to focus on the anatomically relevant regions, and has shown to be effective for whole heart segmentation. In the MM-WHS challenge the method of Payer_2018_MMWHS ranked 1st. A summary of the comparison between the segmentation accuracy of the methods evaluated on MM-WHS dataset is presented in Table 4. For more details, please refer to Zhuang_2019_MedIA.
|Ye_2019_Access||94.4/ -||89.5/ -||91.6/ -||87.8/ -||88.9/ -||96.7/ -||86.2/ -||90.7/ -|
|Xu_2018_arXiv||87.9/ -||90.2/ -||83.2/ -||84.4/ -||82.2/ -||91.3/ -||82.1/ -||85.9/ -|
Multi-view CNNs: Another line of research utilizes the volumetric information of the heart by training multi-planar CNNs (axial, sagittal, and coronal views) in a 2D fashion. Examples include Wang_2017_MMWHS and Mortazi_2017_MMWHS where three independent orthogonal CNNs were trained to segment different views. Specifically, Wang_2017_MMWHS additionally incorporated shape context in the framework for the segmentation refinement, while Mortazi_2017_MMWHS adopted an adaptive fusion strategy to combine multiple outputs utilising complementary information from different planes.
Hybrid loss: Several methods employ a hybrid loss, where different loss functions (such as focal loss, Dice loss, and weighted categorical cross-entropy) are combined to address the class imbalance issue, e.g. the volume size imbalance among different ventricular structures, and to improve the segmentation performance (Yang_2017c_MMWHS; Ye_2019_Access).
In addition, the work of Zreik_2018_MedIA has proposed a method for the automatic identification of patients with significant coronary artery stenoses through the segmentation and analysis of the LV myocardium. In this work, a multi-scale FCN is first employed for myocardium segmentation, and then a convolutional autoencoder is used to characterize the LV myocardium, followed by a SVM to classify patients based on the extracted features.
4.2.2 Coronary Artery Segmentation
Quantitative analysis of coronary arteries is an important step for the diagnosis of cardiovascular diseases, stenosis grading, blood flow simulation and surgical planning (Zhang_2010_Thesis). Though this topic has been studied for years (Lesage_2009_MedIA), only a small number of works investigate the use of deep learning in this context. Methods relating to coronary artery segmentation can be mainly divided into two categories: centerline extraction and lumen (i.e. vessel wall) segmentation.
CNNs as a post-/pre-processing step: Coronary centerline extraction is a challenging task due to the presence of nearby cardiac structures and coronary veins as well as motion artifacts in cardiac CT. Several deep learning approaches employ CNNs as either a post-processing or pre-processing step for traditional methods. For instance, Gulsun_2016_MICCAI formulated centerline extraction as finding the maximum flow paths in a steady state porous media flow, with a learning-based classifier estimating anisotropic vessel orientation tensors for flow computation. A CNN classifier was then employed to distinguish true coronary centerlines from leaks into non-coronary structures. Guo_2019_IPMI proposed a multi-task FCN centerline extraction method that can generate a single-pixel-wide centerline, where the FCN simultaneously predicted centerline distance maps and endpoint confidence maps from coronary arteries and ascending aorta segmentation masks, which were then used as input to the subsequent minimal path extractor to obtain the final centerline extraction results. In contrast, unlike the aforementioned methods that used CNNs either as a pre-processing or post-processing step, Wolterink_2019_MedIA
proposed to address centerline extraction via a 3D dilated CNN, where the CNN was trained on patches to directly determine a posterior probability distribution over a discrete set of possible directions as well as to estimate the radius of an artery at the given point.
End-to-end CNNs: With respect to the lumen or vessel wall segmentation, most deep learning based approaches use an end-to-end CNN segmentation scheme to predict dense segmentation probability maps (Moeskops_2016_MICCAI; Merkow_2016_MICCAI; Huang_2018_EMBC; Shen_2019_IEEEAccess). In particular, Moeskops_2016_MICCAI proposed a multi-task segmentation framework where a single CNN can be trained to perform three different tasks including coronary artery segmentation in cardiac CTA and tissue segmentation in brain MR images. They showed that such a multi-task segmentation network in multiple modalities can achieve equivalent performance as a single task network. Merkow_2016_MICCAI introduced deep multi-scale supervision into a 3D U-net architecture, enabling efficient multi-scale feature learning and precise voxel-level predictions. Besides, shape priors can also be incorporated into the network (Lee_2019_TMI; Chen_2019_MIDL; Duan_2018_STACOM). For instance, Lee_2019_TMI explicitly enforced a roughly tubular shape prior for the vessel segments by introducing a template transformer network, through which a shape template can be deformed via network-based registration to produce an accurate segmentation of the input image, as well as to guarantee topological constraints. More recently, graph convolutional networks have also been investigated by Wolterink_2019_arXiv for coronary artery segmentation in CTA, where vertices on the coronary lumen surface mesh were considered as graph nodes and the locations of these tubular surface mesh vertices were directly optimized. They showed that such method significantly outperformed a baseline network that used only fully-connected layers on healthy subjects (mean Dice score: 0.75 vs 0.67 ). Besides, the graph convolutional network used in their work is able to directly generate smooth surface meshes without post-processing steps.
4.2.3 Coronary Artery Calcium and Plaque Segmentation
CAC is a direct risk factor for cardiovascular disease. Clinically, CAC is quantified using the Agatston score (Agatston_1990_JACC) which considers the lesion area and the weighted maximum density of the lesion (deVos_2019_TMI). Precise detection and segmentation of CAC are thus important for the accurate prediction of the Agatston score and disease diagnosis.
Two-step segmentation: One group of deep learning approaches to segmentation and automatic calcium scoring proposed to use a two-step segmentation scheme. For example, Wolterink_2016_MedIA attempted to classify CAC in cardiac CTA using a pair of CNNs, where the first CNN coarsely identified voxels likely to be CAC within a ROI detected using (De_2017_TMI) and then the second CNN further distinguished between CAC and CAC-like negatives more accurately. Similar to such a two-stage scheme, Lessmann_2016_MedicalImaging; Lessmann_2017_TMI proposed to identify CAC in low-dose chest CT, in which a ROI of the heart or potential calcifications were first localized followed by a CAC classification process.
Direct segmentation: More recently, several approaches (Shadmi_2018_ISBI; Santini_2017_IFMBE; Ma_2019_arXiv; Zhang_2019_Computing) have been proposed for the direct segmentation of CAC from non-contrast cardiac CT or chest CT: the majority of them employed combinations of U-net (Ronneberger_2015_MICCAI) and DenseNet (Huang_2017_CVPR) for precise quantification of CAC which showed that a sensitivity over 90% can be achieved Santini_2017_IFMBE. These aforementioned approaches all follow the same workflow where the CAC is first identified and then quantified. An alternative approach is to circumvent the intermediate segmentation and to perform direct quantification, such as in (Cano_2018_MedicalImaging; deVos_2019_TMI), which have proven that this approach is effective and promising.
Finally, for NCP and MCP in coronary arteries, only a limited number of works have been reported that investigate deep learning methods for segmentation and quantification (Zreik_2018_TMI; Liu_2018_STACOM). Yet, this is a very important task from a clinical point of view, since these plaques can potentially rupture and obstruct an artery, causing ischemic events and severe cardiac damage. In contrast to CAC segmentation, NCP and MCP segmentation are more challenging due to their similar appearances and intensities as adjacent tissues. Therefore, robust and and accurate analysis often requires the generation of MPR images that have been straightened along the centreline of the vesssel. Recently, Liu_2018_STACOM proposed a vessel-focused 3D convolutional network with attention layers to segment three types of plaques on the extracted and reformatted coronary MPR volumes. Zreik_2018_TMI presented an automatic method for detection and characterization of coronary artery plaques as well as determination of coronary artery stenosis significance, in which a multi-task convolutional RNN was used to perform both plaque and stenosis classification by analyzing the features extracted along the coronary artery in an MPR image.
4.3 Cardiac Ultrasound Image Segmentation
|Application||Selected works||Method||Structure||Imaging modality|
|Combined with deformable models|
|LV||2D A2C, A4C|
|LV||2D A2C, A4C|
|LV||2D A2C, A4C|
|Utilizing temporal coherence|
|LV||2D A2C, A4C|
|Utilizing unlabeled data|
|LV||2D A2C, A4C|
|LV and LA||2D A2C, A4C|
|Utilizing data from multiple domains|
|Chen_2016_MICCAI||FCN trained using annotated data of multiple anatomical structures||Fetal head and LV||2D head, A2-5C|
|Trained directly on large datasets|
|LV||2D A2C, A4C|
|Oktay_2018_TMI||U-net with TL-net based shape constraint||LV||3D (CETUS)|
|Multi-chamber||2D PLAX, PSAX, A2-4C|
Cardiac ultrasound (US) imaging, also known as echocardiography, is an indispensable clinical tool for the assessment of cardiovascular function. It is often used clinically as the first imaging examination owing to its portability, low cost and real-time capability. While a number of traditional methods such as active contours, level-sets and active shape models have been employed to automate the segmentation of anatomical structures in ultrasound images (Noble_2006_TMI)
, the achieved accuracy is limited by various problems of ultrasound imaging such as low signal-to-noise ratio, varying speckle noise, low image contrast (especially between the myocardium and the blood pool), edge dropout and shadows cast by structures such as dense muscle and ribs.
As in cardiac MR and CT, several DL-based methods have been recently proposed to improve the performance of cardiac ultrasound image segmentation in terms of both accuracy and speed. The majority of these DL-based approaches focus on LV segmentation, with only few addressing the problem of aortic valve and LA segmentation. A summary of the reviewed works can be found in Table 5.
4.3.1 2D LV segmentation
Deep learning combined with deformable models: The imaging quality of echocardiography makes voxel-wise tissue classification highly challenging. To address this challenge, deep learning has been combined with deformable model for LV segmentation in 2D images (Carneiro_2010_ISBI; Carneiro_2012_TIP; Carneiro_2010_CVPR; Carneiro_2013_TPAMI; Nascimento_2014_CVPR; Nascimento_2019_TPAMI; Veni_2018_ISBI). Features extracted by trained deep neural networks were used instead of handcrafted features to improve accuracy and robustness.
Several works applied deep learning in a two-stage pipeline which first localizes the target ROI via rigid transformation of a bounding box, then segments the target structure within the ROI. This two-stage pipeline reduces the search region of the segmentation and increases robustness of the overall segmentation framework. Carneiro_2010_ISBI; Carneiro_2012_TIP first adopted this DL framework to segment the LV in apical long-axis echocardiograms. The method uses DBN (Hinton_2006_Science) to predict the rigid transformation parameters for localization and the deformable model parameters for segmentation. The results demonstrated the robustness of DBN-based feature extraction to image appearance variations. Nascimento_2017_TIP further reduced the training and inference complexity of the DBN-based framework by using sparse manifold learning in the rigid detection step.
To further reduce the computational complexity, some works perform segmentation in one step without resorting to the two-stage approach. Nascimento_2014_CVPR; Nascimento_2019_TPAMI applied sparse manifold learning in segmentation, showing a reduced training and search complexity compared to their previous version of the method, while maintaining the same level of segmentation accuracy. Veni_2018_ISBI applied a FCN to produce coarse segmentation masks, which is then further refined by a level-set based method.
Utilizing temporal coherence: Cardiac ultrasound data is often recorded as a temporal sequence of images. Several approaches aim to leverage the coherence between temporally close frames to improve the accuracy and robustness of the LV segmentation. Carneiro_2010_CVPR; Carneiro_2013_TPAMI proposed a dynamic modeling method based on a SMC (or particle filtering) framework with a transition model, in which the segmentation of the current cardiac phase depends on previous phases. The results show that this approach performs better than the previous method (Carneiro_2010_ISBI) which does not take temporal information into account. In a more recent work, Jafari_2018_DLMIA combined U-net, LSTM and inter-frame optical flow to utilize multiple frames for segmenting one target frame, demonstrating improvement in overall segmentation accuracy. The method was also shown to be more robust to image quality variations in a sequence than single-frame U-net.
Utilizing unlabeled data: Several works proposed to use non-DL based segmentation algorithms to help generating labels on unlabeled images, effectively increasing the amount of training data. To achieve this, Carneiro_2011_ICCV; Carneiro_2012_CVPR proposed on-line retraining strategies where segmentation network (DBN) is firstly initialized using a small set of labeled data and then applied to non-labeled data to propose annotations. The proposed annotations are then checked by external classifiers before being used to re-train the network. Smistad_2017_IUS trained a U-net using images annotated by a Kalman filtering based method (Smistad_2014_MIDAS) and illustrated the potential of using this strategy for pre-training. Alternatively, some works proposed to exploit unlabeled data without using additional segmentation algorithm. Yu_2017_BiomedEng proposed to train a CNN on a partially labeled dataset of multiple sequences, then fine-tuned the network for each individual sequence using manual segmentation of the first frame as well as CNN-produced label of other frames. Jafari_2019_ISBI proposed a semi-supervised framework which enables training on both the labeled and unlabeled images. The framework uses an additional generative network, which is trained to generate ultrasound images from segmentation masks, as additional supervision for the unlabeled frames in the sequences. The generative network forces the segmentation network to predict segmentation that can be used to successfully generate the input ultrasound image.
Utilizing data from multiple domains: Apart from exploiting unlabeled data in the same domain, leveraging manually annotated data from multiple domains (e.g. different 2D ultrasound views with various anatomical structures) can also help to improve the segmentation in one particular domain. Chen_2016_MICCAI proposed a novel FCN-based network to utilize multi-domain data to learn generic feature representations. Combined with an iterative refinement scheme, the method has shown superior performance in detection and segmentation over traditional database-guided method (Georgescu_2005_CVPR), FCN trained on single-domain and other multi-domain training strategies.
DL networks trained directly on large datasets: The potential of CNN in segmentation has motivated the collection and labeling of large-scale datasets. Several methods have since shown that deep learning methods, most notably CNN-based methods, are capable of performing accurate segmentation directly without complex post-processing. Leclerc_2018_IUS performed a study to investigate the effect of the size of annotated data for the segmentation of the LV in 2D ultrasound images using a simple U-net. The authors demonstrated that the U-net approach significantly benefits from larger amounts of training data. Furthermore, Smistad_2018_IUS demonstrated the efficiency of CNN-based methods by successfully performing real-time view-classification and segmentation.
4.3.2 3D LV segmentation
Segmenting cardiac structures in 3D ultrasound is even more challenging than 2D. While having the potential to derive more accurate volume-related clinical indices, 3D echocardiograms suffer from lower temporal resolution and lower image quality compared to 2D echocardiograms. Moreover, 3D images dramatically increase the dimension of parameter space of neural networks, which poses computational challenges for deep learning methods.
One way to reduce the computational cost is to avoid direct processing of 3D data in deep learning networks. Dong_2018_HindawiBiomed proposed a two-stage method by first applying a 2D CNN to produce coarse segmentation maps on 2D slices from a 3D volume. The coarse 2D segmentation maps are used to initialize a 3D shape model which is then refined by 3D deformable model method (Kass_1988_IJCV). In addition, the authors used transfer learning to side-step the limited training data problem by pre-training network on a large natural image segmentation dataset and then fine-tuning to the LV segmentation task.
Anatomical shape priors have been utilized to increase the robustness of deep learning-based segmentation methods to challenging 3D ultrasound images. Oktay_2018_TMI proposed an anatomically constrained network where a shape constraint-based loss is introduced to train a 3D segmentation network. The shape constraint is based on the shape prior learned from segmentation maps using auto-encoders (Girdhar_2016_ECCV). Dong_2018_MICCAI_echo utilized shape prior more explicitly by combining a neural network with a conventional atlas-based segmentation framework. Adversarial training was also applied to encourage the method to produce more anatomically plausible segmentation maps, which contributes to its superior segmentation performance comparing to a standard voxel-wise classification 3D segmentation network (Milletari_2016_3DV).
4.3.3 Left-atrium segmentation
Degel_2018_MICCAI adopted the aforementioned anatomical constrain in 3D LA segmentation to tackle the domain shift problem caused by variation of imaging device, protocol and patient condition. In addition to the anatomically constraining network, the authors applied an adversarial training scheme (Kamnitsas_2017_IPMI) to improve the generalizability of the model to unseen domain.
4.3.4 Multi-chamber segmentation
Apart from LV segmentation, a few works (Zhang_2018_Circulation; Smistad_2017_IUS; Leclerc_2019_TMI) applied deep learning methods to perform multi-chamber (including LV and LA) segmentation. In particular, Zhang_2018_Circulation demonstrated the applicability of CNNs on three tasks: view classification, multi-chamber segmentation and detection of cardiovascular diseases. Comprehensive validation on a large (non-public) clinical dataset showed that clinical metrics derived from automatic segmentation are comparable or superior than manual segmentation. To resemble real clinical situations and thus encourages the development and evaluation of robust and clinically effective segmentation methods, a large-scale dataset for 2D cardiac ultrasound has been recently made public(Leclerc_2019_TMI). The dataset and evaluation platform were released following the preliminary data requirement investigation of deep learning methods (Leclerc_2018_IUS). The dataset is composed of apical 4-chamber view images annotated for LV and LA segmentation, with uneven imaging quality from 500 patients with varying conditions. Notably, the initial benchmarking (Leclerc_2019_TMI) on this dataset has shown that modern encoder-decoder CNNs resulted in lower error than inter-observer error between human cardiologists.
4.3.5 Aortic valve segmentation
Ghesu_2016_TMI proposed a framework based on MSL, DNN and ASM to segment the aortic valve in 3D cardiac ultrasound volumes. An adaptive sparsely-connected neural network with reduced number of parameters is used to predict a bounding box to locate the target structure, where the learning of the bounding box parameters is marginalized into sub-spaces to reduce computational complexity. This framework showed significant improvement over the previous non-DL MSL (Zheng_2008_TMI) method while achieving competitive run-time.
So far, we have presented and discussed recent progress of deep learning-based segmentation methods in the three modalities (i.e. MR, CT, US) that are commonly used in the assessment of cardiovascular disease. To summarize, current state-of-the-art segmentation methods are mainly based on CNNs that employ the FCN or U-net architecture. In addition, there are several commonalities in the FCN-based methods for cardiac segmentation which can be categorized into four groups: 1) enhancing network feature learning by employing advanced building blocks in networks (e.g. inception module, dilated convolutions), most of which have been mentioned earlier (Sec. 3.1.5); 2) alleviating the problem of class imbalance with advanced loss functions (e.g. weighted loss functions); 3) improving the networks’ generalization ability and robustness through a multi-stage pipeline, multi-task learning, or multi-view feature fusion; 4) forcing the network to generate more anatomically-plausible segmentation results by incorporating shape priors, applying adversarial loss or anatomical constraints to regularize the network during training. It is also worthwhile to highlight that for cardiac image sequence segmentation (e.g. cine MR images, 2D US sequences), leveraging spatial and temporal coherence from these sequences with advanced neural networks (e.g. RNN (Bai_2018_MICCAI; Jafari_2018_DLMIA), multi-slice FCN (Zheng_2018_TMI)) has been explored and shown to be beneficial for improving the segmentation accuracy and temporal consistency of the segmentation maps.
While the results reported in the literature show that neural networks have become more sophisticated and powerful, it is also clear that performance has improved with the increase of publicly available training subjects. A number of DL-based methods (especially in MRI) have been trained and tested on public challenge datasets, which not only provide large amounts of data to exploit the capabilities of deep learning in this domain, but also a platform for transparent evaluation and comparison. In addition, many of the participants in these challenges have shared their code with other researchers via open-source community websites (e.g. Github). Transparent and fair benchmarking and sharing of code are both essential for continued progress in this domain. We summarize the existing public datasets in Table6 and public code repositories in Table 7 for reference.
|Dataset Name/Reference||Year||Main modalities||#||Target(s)||Main Pathology|
|York (Andreopoulos_2008_York)||2008||cine MRI||33||LV||cardiomyopathy, aortic regurgitation, enlarged ventricles and ischemia|
|Sunnybrook (Radau_2009_Sunnybrook)||2009||cine MRI||45||LV||hypertrophy, heart failure w./w.o infarction|
|LVSC (Suinesiaputra_LVSC_2011)||2011||cine MRI||200||LV||coronary artery disease, myocardial infarction.|
|RVSC (Petitjean_2015_MedIA)||2012||cine MRI||48||RV||
|cDEMRIS (Karim_2013_LA_SCAR)||2012||LGE MRI||60||LA fibrosis and scar||atrial fibrillation|
|LVIC (Karim_2016_LVIS_Dataset)||2012||LGE MRI||30||Myocardial scars||ischaemic cardiomyopathy|
|LASC’13 (Tobon-Gomez_2015_LASC)||2013||3D MRI||30||LA||N/A|
|HVSMR (Pace_2015_HVSMR)||2016||3D MRI||4||Blood pool, MYO||congenital heart defects|
|ACDC (Bernard_2018_TMI)||2017||MRI||150||LV; RV||mycardial infarction, dilated/ hypertrophic cardiomyopathy, abnormal RV|
|LASC’18 (LASC_2018)||2018||LGE MRI||150||LA||atrial fibrillation|
|CAT08 (Schaap_2009_centerline_dataset)||2008||CTA||32||Coronary artery centerline||Patients with presence of calcium scored as absent, modest or severe.|
|CLS12 (Kirisli2013_Lumen_Stenosis_dataset)||2012||CTA||48||Coronary lumen and stenosis||Patients with different levels of coronary artery stenoses.|
|CETUS (CETUS_2016)||2014||3D US||45||LV||myocardial infarction, dilated cardiomyopathy|
|CAMUS (Leclerc_2019_TMI)||2019||2D US||500||LV, LA||Patients with EF|
|Modality||Application(s)||Authors||Basic Network||Code Repo (If not specified, the repository is located under github.com)|
|MR (SAX)||Bi-ventricular Segmentation||Tran_2016_Arxiv||2D FCN||vuptran/cardiac-segmentation|
|MR (SAX)||Bi-ventricular Segmentation||Baumgartner_2017_STACOM||2D/3D U-net||baumgach/acdc_segmenter|
|MR (SAX)||Bi-ventricular Segmentation; 1st rank in ACDC challenge||Isensee_2017_STACOM||2D+3D U-net (ensemble)||MIC-DKFZ/ACDC2017|
|MR (SAX)||Bi-ventricular Segmentation||Zheng_2018_TMI||cascaded 2D U-net||julien-zheng/CardiacSegmentationPropagation|
|MR (SAX)||Bi-ventricular segmentation and Motion Estimation||Qin_2018_MICCAI||2D FCN, RNN||cq615|
|MR (SAX)||Biventricular Segmentation||Khened_2019_MedIA||2D U-net||mahendrakhened|
|MR (3D scans)||Blood pool+MYO Segmentation||Yu_2017_MICCAI||3D CNN||yulequan/HeartSeg|
|MR (Multi-view)||Four-chamber Segmentation and Aorta Segmentation||Bai_2018_JCMR; Bai_2018_MICCAI||2D FCN, RNN||baiwenjia/ukbb_cardiac|
|MR||Cardiac Segmentation and Motion Tracking||Duan_2019_TMI||2.5D FCN +Atlas-based||j-duan/4Dsegment|
|LGE MRI||Left Atrial Segmentation||Chen_2018_STACOM||2D U-net||cherise215/atria_segmentation_2018|
|LGE MRI||Left Atrial Segmentation||Yu_2019_MICCAI||3D V-net||yulequan/UA-MT|
|CT||WHS||Xu_2018_arXiv||Faster RCNN, 3D U-net||Wuziyi616/CFUN|
|CT, MRI||Coronary arteries||Merkow_2016_MICCAI||3D U-net||jmerkow/I2I|
|CT, MRI||WHS||Dou_2018_IJCAI; Dou_2019_Access||2D CNN||carrenD/Medical-Cross-Modality-Domain-Adaptation|
|CT, MRI||WHS||Chen_2019_AAAI||2D CNN||cchen-cc/SIFA|
|US||View Classification and Four-chamber Segmentation||Zhang_2018_Circulation||2D U-net||bitbucket.org/rahuldeo/echocv|
2D Networks vs 3D Networks: An interesting conclusion supported by Table 7 is that the target image type can affect the choice of network structures (i.e. 2D networks, 3D networks). For 3D imaging acquisitions such as LGE-MRI and CT images, 3D networks are preferred whereas 2D networks are more popular approaches for segmenting cardiac cine short-axis or long-axis image stacks. One reason for using 2D networks for the segmentation of short-axis or long-axis images is their typically large slice thickness (usually around 7–8 mm) which can further exacerbated by inter-slice gaps. In addition, breath-hold related motion artifacts between different slices may negatively affect 3D networks. A study conducted by Baumgartner_2017_STACOM has shown that a 3D U-net performs worse than a 2D U-net when evaluated on the ACDC challenge dataset. By contrast, in the LASC’18 challenge mentioned in Table 6, which uses high-resolution 3D images, most participants applied 3D networks and the best performance was achieved by a cascaded network based on the 3D U-net (Xia_2018_STACOM).
It is well known that training 3D networks is more difficult than training 2D networks. In general, 3D networks have significantly more parameters than 2D networks. Therefore, 3D networks are more difficult and computationally expensive to optimize as well as prone to over-fitting, especially if the training data is limited. As a result, several researchers have tried to carefully design the structure of network to reduce the number of parameters for a particular application and have also applied advanced techniques (e.g. deep supervision) to alleviate the over-fitting problem (Yu_2017_MICCAI; Xia_2018_STACOM). For this reason, 2D-based networks (e.g. 2D U-net) are still the most popular segmentation approaches for all three modalities.
In addition to 2D and 3D networks, several authors have proposed ‘2D+’ networks that have been shown to be effective in segmenting structures from cardiac volumetric data. These ‘2D+’ networks are mainly based on 2D networks, but are adapted with increased capacity to utilize 3D context. These networks include multi-view networks which leverage multi-planar information (i.e. coronal, sagittal, axial views) (Mortazi_2017_STACOM; Wang_2017_MMWHS), multi-slice networks, and 2D FCNs combined with RNNs which incorporate context across multiple slices (Duan_2019_TMI; Patravali_2017_STACOM; Poudel_2016_HVSCMR; Du_2019_JTEHM). These ‘2D+’networks inherit the advantages of 2D networks while still being capable of leveraging through-plane spatial context for more robust segmentation with strong 3D consistency.
5 Challenges and Future Work
It is evident from the literature that deep learning methods have matched or surpassed the previous state of the art in a various cardiac segmentation applications, mainly benefiting from the increased size of public datasets and the emergence of advanced network architectures as well as powerful hardware for computing. Given this rapid process, one may wonder if deep learning methods can be directly deployed to real-world applications to reduce the workload of clinicians. The current literature suggests that there is still a long way to go. In the following paragraphs, we summarize several major challenges in the field of cardiac segmentation and some recently proposed approaches that attempt to address them. These challenges and related works also provide potential research directions for future work in this field.
5.1 Scarcity of Labels
One of the biggest challenges for deep learning approaches is the scarcity of annotated data. In this review, we found that the majority of studies uses a fully supervised approach to train their networks, which requires a large number of annotated images. In fact, annotating cardiac images is time consuming and often requires significant amounts of expertise. While data augmentation techniques such as cropping, padding, and geometric transformations (e.g. affine transformations) can be used to increase the size of training samples, their diversity may still be limited, failing to reflect the spectrum of real-world data distributions. Several methods have been proposed to overcome this challenge. These methods can be categorized into four classes: transfer learning with fine-tuning, weakly and semi-supervised learning, self-supervised learning, and unsupervised learning.
Transfer learning with fine-tuning. Transfer learning aims at reusing a model pre-trained on one task as a starting point to train for a second task. The key of transfer learning is to learn features in the first task that are related to the second task such that the network can quickly converge even with limited data. Several researchers have successfully demonstrated the use of transfer learning to improve the model generalization ability for cardiac ventricle segmentation across different scanners, where they first trained a model on a large dataset and then fine-tuned it on a small dataset (Bai_2018_JCMR; Khened_2019_MedIA; Cong_2018_JE; Fahmy_2019_JCMR; chen_2019_med3d).
Weakly and semi-supervised learning. Weakly and semi-supervised learning methods aim at improving the learning accuracy by making use of both labeled and unlabeled or weakly-labeled data (e.g annotations in forms of scribbles or bounding boxes). In this context, several works have been proposed for cardiac ventricle segmentation in MR images. One approach is to estimate full labels on unlabeled or weakly labeled images for further training. For example, Bai_2018_MICCAI; Qin_2018_MICCAI utilized motion information to propagate labels from labeled frames to unlabeled frames in a cardiac cycle whereas Bai_2017_MICCAI; Can_2018_DLMIAML applied the EM algorithm to predict and refine the estimated labels recursively. Others have explored different approaches to regularize the network when training on unlabeled images, applying multi-task learning (ChartsiasA_2018), or global constraints (Kervadec_2018_MeDIA).
Self-supervised learning. Another approach is self-supervised learning which aims at utilizing labels that are generated automatically without human intervention. These labels, designed to encode some properties or semantics of the object, can provide strong supervisory signals to pre-train a network before fine-tuning for a given task. A very recent work from Bai_2019_MICCAI has shown the effectiveness of self-supervised learning for cardiac MR image segmentation where the authors used auto-generated anatomical position labels to pre-train a segmentation network. Compared to a network trained from scratch, networks pre-trained on the self-supervised task performed better, especially when the training data was extremely limited.
Unsupervised learning. Unsupervised learning aims at learning without paired labeled data. Compared to the former three classes, there is limited literature about unsupervised learning methods for cardiac image segmentation, perhaps because of the difficulty of the task. An early attempt has been made which applied adversarial training to train a network segmenting LV and RV from CT and MR images without requiring a training set of paired images and labels (Joyce_2018_MIDL).
Apart from utilizing unlabeled images for training neural networks, another interesting direction is active learning(Mahapatra_2018_MICCAI), which tries to select the most representative images from a large-scale dataset, reducing redundant labeling workload and training cost. This technique is also related to incremental learning, which aims to improve the model performance with new classes added incrementally while avoiding a dramatic decrease in overall performance (Castro_2018_ECCV). Given the increasing size of the available medical datasets, and the practical challenges of labeling and storing large amounts of images from various sources, it is of great interest to develop algorithms capable of distilling a large-scale cardiac dataset into a small one containing the most representative cases for labeling and training.
5.2 Model Generalization Across Various Imaging Modalities, Scanners and Pathologies.
Another common limitation in DL-based methods is that they still lack generalization capabilities when presented with previously unseen samples (e.g. data from a new scanner, abnormal and pathological cases that have not been included in the training set). In other words, deep learning models tend to be biased by their respective training datasets. This limitation prevents models to be deployed in the real world and therefore diminishes their impact for improving clinical workflows. To improve the model performance across MR images acquired from multiple vendors and multiple scanners, Tao_2019_Radiology collected a large multi-vendor, multi-center, heterogeneous labeled training set from patients with cardiovascular diseases. However, this approach may not scale to the real world, as it implies the collection of a vastly large dataset covering all possible cases. Moreover, it still faces the aforementioned collecting and labeling challenge.
Unsupervised domain adaptation. Several researchers have recently started to investigate the use of unsupervised domain adaptation techniques that aim at optimizing the model performance on unseen datasets without additional labeling costs. Several works have successfully applied adversarial training for cross-modality segmentation tasks, adapting a cardiac segmentation model learned from MR images to CT images and vice versa (Dou_2018_IJCAI; Dou_2019_Access; Ouyang_2019_MICCAI; Chen_2019_AAAI). These type of approaches can also be adopted for semi-supervised learning, where the target domain is a new set of unlabeled data of the same modality (Chen_2019_MICCAI_UDA).
Data augmentation. An alternative yet simple and effective approach is data augmentation. The main idea is to increase the variety of training images so that the training set distribution is more close to the one of a test set in the real world. In general, this type of augmentation is achieved by applying a stack of geometric or photometric transformations to existing image-label pairs. Recently, Chen_2019_Arxiv have proposed a data normalization and augmentation pipeline which enables a neural network for cardiac MR image segmentation trained from a single-scanner dataset to generalize well across multi-scanner and multi-site datasets. Zhang_2019_Arxiv applied a similar data augmentation approach to improve the model generalization ability on unseen datasets. Their method has been verified on three tasks including left atrial segmentation from 3D MRI and left ventricle segmentation from 3D ultrasound images. However, effectively designing such a pipeline requires expertise, which may not be easy to be extended to other applications. Most recently, several researchers have began to investigate the use of generative models (e.g. GAN, variational AE (Kingma_2013_ICLR)Ekin_2019_CVPR) and adversarial example generation (Volpi_2018_NIPS) that aim at directly learning data augmentation strategies from existing data. In particular, the generative model-based approach has been proven to be effective for one-shot brain segmentation (Zhao_2019_CVPR) and few-shot cardiac MR image segmentation (Chaitanya_2019_IPMI) and is thus worth to be explored for more applications in the future.
5.3 Lack of Model Interpretability
Unlike symbolic artificial intelligence systems, deep learning systems are difficult to interpret and not transparent. Once a network has been trained, it behaves like a ‘black box’, providing predictions which are not directly interpretable. This issue makes the model unpredictable, intractable for model verification, and ultimately untrustworthy. Recent studies have shown that deep learning-based vision recognition systems can be attacked by images modified with nearly imperceptible perturbations (Szegedy_2013_Arxiv; Kurakin_2016_Arxiv; Goodfellow_2015_ICLR). These attacks can also happen in medical scenarios, e.g. a DL-based system may make a wrong diagnosis given an image with adversarial noise or even just small rotation, as demonstrated in a very recent paper (Finlayson_2019_Science). Although there is no denying that deep learning has become a very powerful tool for image analysis, building resilient algorithms robust to potential attacks remains an unsolved problem. One potential solution, instead of building the resilience into the model, is raising failure awareness of the deployed networks. This can be achieved by providing users with segmentation quality scores (Robinson_2019_JCMR) or confidence maps such as uncertainty maps (Sander_2019_MIP) and attention maps (Heo_2018_NIPS). These scores or maps can be used as evidence to alert users when failure happens. For example, Sander_2019_MIP built a network that is able to simultaneously predict the segmentation mask over cardiac structures and its associated spatial uncertainty map, where the latter one could be used to highlight potential incorrect regions. Such uncertainty information could alert human experts for further justification and refinement in a human-in-the-loop setting.
5.4 Future work
Smart imaging. We have shown that deep learning-based methods are able to segment images in real-time with good accuracy. However, these algorithms can still fail on those image acquisitions with low image quality or significant artifacts. Although there have been several algorithms developed to avoid this problem by either checking the image quality before follow-up studies (Ruijsink_2019_JACC; Tarroni_2019_TMI), or predicting the segmentation quality to detect failures (Peng_2012_ECCV; Robinson_2019_JCMR; Zhou_2019_Arxiv), the development of algorithms that can give instant feedback to correct and optimize the image acquisition process is also important despite less explored. Improving the imaging quality can greatly improve the effectiveness of medical imaging as well as the accuracy of imaging-based diagnosis. For radiologists, however, finding the optimal imaging and reconstruction parameters to scan each patient can take a great amount of time. Therefore, a DL-based system that has the potential of efficiently and effectively improving the image quality with less noise is of great need. Some researchers have utilized learning-based methods (mostly are deep learning-based) for better image resolution (Oktay_2016_MICCAI), view planning (Alansary_2018_MICCAI), motion correction (Dangi_2018_SPIE; Tarroni_2018_MICCAI), artifacts reduction (Ilkay_2019_MIDL), shadow detection (Meng_2019_TMI) and noise reduction (Wolterink_2017_TMI) after image acquisition. However, combining these algorithms with segmentation algorithms and seamlessly integrating them into an efficient, patient-specific imaging system for high-quality image analysis and diagnosis is still an open challenge. An alternative approach is to directly predict cardiac segmentation maps from undersampled k-space data to accelerate the whole procedure, which bypasses the image reconstruction stage (Schlemper_2018_MICCAI).
Data harmonization. A number of works have reported the existence of missing labels and inconsistent labeling protocols among different cardiac image datasets (Zheng_2018_TMI; Chen_2019_Arxiv). Variations have been found in defining the end of basal slices as well as the endocardial wall of myocardium (some include papillary muscles as part of the endocardial contours whereas others do not). These inconsistencies can be a major obstacle for transferring, evaluating and deploying deep learning models trained from one domain (e.g. hospital) to another. Therefore, building a standard benchmark dataset like CheXpert (Irvin_2019_AAAI) that 1) is large enough to have substantial data diversity that reflects the spectrum of real-world diversity; 2) has a standard labeling protocol approved by experts, is indeed a need. However, directly building such a dataset from scratch is time-consuming and expensive. A more promising way might be developing an automated tool to combine existing datasets from multiple sources and then to harmonize them to a unified, high-quality dataset. This tool can not only open the door for crowd-sourcing but also enable the rapid deployment of those DL-based segmentation models.
Data privacy. As deep learning is a data-driven approach, an unavoidable and rife concern is about the data privacy. Regulations such as GDPR now play an important role to protect users’ privacy and have forced organizations to treat data ownership seriously. On the other hand, from a technical point of view, how to store, query, and process data such that there is no privacy concerns for building deep learning systems has now become an even more difficult but interesting challenge. Building a privacy-preserving algorithm requires to combine cryptography and deep learning together and to mix techniques from a wide range of subjects such as data analysis, distributed computing, federated learning, differential privacy, in order to achieve models with strong security, fast run time, and great generalizability (Dwork_2014_FTTCS; Abadi_2016_CCS; Bonawitz_2017_CCS; Ryffel_2018_PPML). In this respect, Papernot_2018_Arxiv published a report for guidance, which summarized a set of best practices for improving the privacy and security of machine learning systems. Yet, this field is still in its infancy.
In this review paper, we provided a comprehensive overview of these deep learning techniques used in three common imaging modalities (MRI, CT, US), covering a wide range of existing deep learning approaches (mostly are CNN-based) that are designed for segmenting different cardiac anatomical structures (e.g. cardiac ventricle, atria, vessel). In particular, we presented and discussed recent progress of deep learning-based segmentation methods in the three modalities, outlined future potential and the remaining limitations of these deep learning-based cardiac segmentation methods that may hinder widespread clinical deployment. We hope that this review can provide an intuitive understanding of those deep learning-based techniques that have made a significant contribution to cardiac image segmentation and also increase the awareness of common challenges in this field that call for future contribution.
CT: computed tomography;
CTA: computed tomography angiography;
MPR: multi-planar reformatted;
MR: magnetic resonance;
MRI: magnetic resonance imaging;
LGE: late gadolinium enhancement;
RFCA: radio-frequency catheter ablation;
Cardiac structures and indexes: AF: atrial fibrillation; AS: aortic stenosis; AO: aorta; CVD: cardiovascular diseases; CAC: coronary artery calcium; DCM: dilated cardiomyopathy; ED: end-diastole; ES: end-systole; EF: ejection fraction; HCM: hypertrophic cardiomyopathy; LA: left atrium; LV: left ventricle; LVEDV: left ventricular end-diastolic volume; LVESV: left ventricular end-systolic volume; MCP: mixed-calcified plaque; MI: myocardial infarction; MYO: myocardium; NCP: non-calcified plaque; PA: pulmonary artery; PV: pulmonary vein; RA: right atrium; RV: right ventricle; RVEDV: right ventricular end-diastolic volume; RVESV: right ventricular end-systolic volume; RVEF: right ventricular ejection fraction; WHS: whole heart segmentation.
Machine learning terminology:
AE: autoencoder; ASM: active shape model; BN: batch normalization; CNN: convolutional neural network; CRF: conditional random field; DBN: deep belief network; DL: deep learning; DNN: deep neural network; EM: expectation maximization; FCN: fully convolutional neural network; GAN: generative adversarial network; GRU: gated recurrent units; MSE: mean squared error; MSL: marginal space learning; MRF: markov random field; LSTM: Long-short term memory; ReLU: rectified linear unit; RNN: recurrent neural network; ROI: region-of-interest; SMC: sequential monte carlo; SRF: structured random forest; SVM: support vector machine.
Cardiac image segmentation datasets: ACDC: Automated Cardiac Diagnosis Challenge; CETUS: Challenge on Endocardial Three-dimensional Ultrasound Segmentation; MM-WHS: Multi-Modality Whole Heart Segmentation; LASC: Left Atrium Segmentation Challenge; LVSC: Left Ventricle Segmentation Challenge; RVSC: Right Ventricle Segmentation Challenge.
Others: EMBC: The International Engineering in Medicine and Biology Conference; GDPR: The General Data Protection Regulation; GPU: graphic processing unit; FDA: United States Food and Drug Administration; ISBI: The IEEE International Symposium on Biomedical Imaging; MICCAI: International Conference on Medical Image Computing and Computer-assisted Intervention; TPU: tensor processing unit; WHO: World Health Organization.
Conflict of Interest Statement
The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.
CC, WB, DR conceived and designed the work; CC, CQ, HQ searched and read the MR, CT, US literature, respectively; CC, CQ, HQ drafted the manuscript together; WB, DR, GT, JD provided critical revision with insightful and constructive comments to improve the manuscript; All authors read and approved the manuscript.
This work is supported by the SmartHeart EPSRC Programme Grant (EP/P001009/1). Huaqi Qiu is supported by the EPSRC Programme Grant (EP/R005982/1).
Data Availability Statement
The datasets summarized in Table 6 can be found in their corresponding websites listed below:
We would like to thank our colleagues: Karl Hahn, Qingjie Meng, James Batten, and Jonathan Passerat-Palmbach who provided insight and expertise that greatly assisted the work, and also constructive and thoughtful comments from Turkay Kart that greatly improved the manuscript.