An efficient framework for lesion segmentation in ultrasound images using global adversarial learning and region-invariant loss.

Lesion segmentation in ultrasound images is an essential yet challenging step for early evaluation and diagnosis of cancers. In recent years, many automatic CNN-based methods have been proposed to assist this task. However, most modern approaches often lack capturing long-range dependencies and prior information making it difficult to identify the lesions with unfixed shapes, sizes, locations, and textures. To address this, we present a novel lesion segmentation framework that guides the model to learn the global information about lesion characteristics and invariant features (e.g., morphological features) of lesions to improve the segmentation in ultrasound images. Specifically, the segmentation model is guided to learn the characteristics of lesions from the global maps using an adversarial learning scheme with a self-attention-based discriminator. We argue that under such a lesion characteristics-based guidance mechanism, the segmentation model gets more clues about the boundaries, shapes, sizes, and positions of lesions and can produce reliable predictions. In addition, as ultrasound lesions have different textures, we embed this prior knowledge into a novel region-invariant loss to constrain the model to focus on invariant features for robust segmentation. We demonstrate our method on one in-house breast ultrasound (BUS) dataset and two public datasets (i.e., breast lesion (BUS B) and thyroid nodule from TNSCUI2020). Experimental results show that our method is specifically suitable for lesion segmentation in ultrasound images and can outperform the state-of-the-art approaches with Dice of 0.931, 0.906, and 0.876, respectively. The proposed method demonstrates that it can provide more important information about the characteristics of lesions for lesion segmentation in ultrasound images, especially for lesions with irregular shapes and small sizes. It can assist the current lesion segmentation models to better suit clinical needs.

Thyroid ultrasound diagnosis improvement via multi-view self-supervised learning and two-stage pre-training.

Thyroid nodule classification and segmentation in ultrasound images are crucial for computer-aided diagnosis; however, they face limitations owing to insufficient labeled data. In this study, we proposed a multi-view contrastive self-supervised method to improve thyroid nodule classification and segmentation performance with limited manual labels. Our method aligns the transverse and longitudinal views of the same nodule, thereby enabling the model to focus more on the nodule area. We designed an adaptive loss function that eliminates the limitations of the paired data. Additionally, we adopted a two-stage pre-training to exploit the pre-training on ImageNet and thyroid ultrasound images. Extensive experiments were conducted on a large-scale dataset collected from multiple centers. The results showed that the proposed method significantly improves nodule classification and segmentation performance with limited manual labels and outperforms state-of-the-art self-supervised methods. The two-stage pre-training also significantly exceeded ImageNet pre-training.

Semi-Supervised Representation Learning for Segmentation on Medical Volumes and Sequences.

Benefiting from the massive labeled samples, deep learning-based segmentation methods have achieved great success for two dimensional natural images. However, it is still a challenging task to segment high dimensional medical volumes and sequences, due to the considerable efforts for clinical expertise to make large scale annotations. Self/semi-supervised learning methods have been shown to improve the performance by exploiting unlabeled data. However, they are still lack of mining local semantic discrimination and exploitation of volume/sequence structures. In this work, we propose a semi-supervised representation learning method with two novel modules to enhance the features in the encoder and decoder, respectively. For the encoder, based on the continuity between slices/frames and the common spatial layout of organs across subjects, we propose an asymmetric network with an attention-guided predictor to enable prediction between feature maps of different slices of unlabeled data. For the decoder, based on the semantic consistency between labeled data and unlabeled data, we introduce a novel semantic contrastive learning to regularize the feature maps in the decoder. The two parts are trained jointly with both labeled and unlabeled volumes/sequences in a semi-supervised manner. When evaluated on three benchmark datasets of medical volumes and sequences, our model outperforms existing methods with a large margin of 7.3% DSC on ACDC, 6.5% on Prostate, and 3.2% on CAMUS when only a few labeled data is available. Further, results on the M&M dataset show that the proposed method yields improvement without using any domain adaption techniques for data from unknown domain. Intensive evaluations reveal the effectiveness of representation mining, and superiority on performance of our method. The code is available at https://github.com/CcchenzJ/BootstrapRepresentation.

RecON: Online learning for sensorless freehand 3D ultrasound reconstruction.

Sensorless freehand 3D ultrasound (US) reconstruction based on deep networks shows promising advantages, such as large field of view, relatively high resolution, low cost, and ease of use. However, existing methods mainly consider vanilla scan strategies with limited inter-frame variations. These methods thus are degraded on complex but routine scan sequences in clinics. In this context, we propose a novel online learning framework for freehand 3D US reconstruction under complex scan strategies with diverse scanning velocities and poses. First, we devise a motion-weighted training loss in training phase to regularize the scan variation frame-by-frame and better mitigate the negative effects of uneven inter-frame velocity. Second, we effectively drive online learning with local-to-global pseudo supervisions. It mines both the frame-level contextual consistency and the path-level similarity constraint to improve the inter-frame transformation estimation. We explore a global adversarial shape before transferring the latent anatomical prior as supervision. Third, we build a feasible differentiable reconstruction approximation to enable the end-to-end optimization of our online learning. Experimental results illustrate that our freehand 3D US reconstruction framework outperformed current methods on two large, simulated datasets and one real dataset. In addition, we applied the proposed framework to clinical scan videos to further validate its effectiveness and generalizability.

Test-time bi-directional adaptation between image and model for robust segmentation.

BACKGROUND AND OBJECTIVE: Deep learning models often suffer from performance degradations when deployed in real clinical environments due to appearance shifts between training and testing images. Most extant methods use training-time adaptation, which almost require target domain samples in the training phase. However, these solutions are limited by the training process and cannot guarantee the accurate prediction of test samples with unforeseen appearance shifts. Further, it is impractical to collect target samples in advance. In this paper, we provide a general method of making existing segmentation models robust to samples with unknown appearance shifts when deployed in daily clinical practice. METHODS: Our proposed test-time bi-directional adaptation framework combines two complementary strategies. First, our image-to-model (I2M) adaptation strategy adapts appearance-agnostic test images to the learned segmentation model using a novel plug-and-play statistical alignment style transfer module during testing. Second, our model-to-image (M2I) adaptation strategy adapts the learned segmentation model to test images with unknown appearance shifts. This strategy applies an augmented self-supervised learning module to fine-tune the learned model with proxy labels that it generates. This innovative procedure can be adaptively constrained using our novel proxy consistency criterion. This complementary I2M and M2I framework demonstrably achieves robust segmentation against unknown appearance shifts using existing deep-learning models. RESULTS: Extensive experiments on 10 datasets containing fetal ultrasound, chest X-ray, and retinal fundus images demonstrate that our proposed method achieves promising robustness and efficiency in segmenting images with unknown appearance shifts. CONCLUSIONS: To address the appearance shift problem in clinically acquired medical images, we provide robust segmentation by using two complementary strategies. Our solution is general and amenable for deployment in clinical settings.

Co-learning of appearance and shape for precise ejection fraction estimation from echocardiographic sequences.

Accurate estimation of ejection fraction (EF) from echocardiography is of great importance for evaluation of cardiac function. It is usually obtained by the Simpson's bi-plane method based on the segmentation of the left ventricle (LV) in two keyframes. However, obtaining accurate EF estimation from echocardiography is challenging due to (1) noisy appearance in ultrasound images, (2) temporal dynamic movement of myocardium, (3) sparse annotation of the full sequence, and (4) potential quality degradation during scanning. In this paper, we propose a multi-task semi-supervised framework, which is denoted as MCLAS, for precise EF estimation from echocardiographic sequences of two cardiac views. Specifically, we first propose a co-learning mechanism to explore the mutual benefits of cardiac segmentation and myocardium tracking iteratively on appearance level and shape level, therefore alleviating the noisy appearance and enforcing the temporal consistency of the segmentation results. This temporal consistency, as shown in our work, is critical for precise EF estimation. Then we propose two auxiliary tasks for the encoder, (1) view classification to help extract the discriminative features of each view, and automatize the whole pipeline of EF estimation in clinical practice, and (2) EF regression to help regularize the spatiotemporal embedding of the echocardiographic sequence. Both two auxiliary tasks can improve the segmentation-based EF prediction, especially for sequences of poor quality. Our method is capable of automating the whole pipeline of EF estimation, from view identification, cardiac structures segmentation to EF calculation. The effectiveness of our method is validated in aspects of segmentation, tracking, consistency analysis, and clinical parameters estimation. When compared with existing methods, our method shows obvious superiority for LV volumes on ED and ES phases, and EF estimation, with Pearson correlation of 0.975, 0.983 and 0.946, respectively. This is a significant improvement for echocardiography-based EF estimation and improves the potential of automated EF estimation in clinical practice. Besides, our method can obtain accurate and temporal-consistent segmentation for the in-between frames, which enables it for cardiac dynamic function evaluation.

Improved Segmentation of Echocardiography With Orientation-Congruency of Optical Flow and Motion-Enhanced Segmentation.

Quantification of left ventricular (LV) ejection fraction (EF) from echocardiography depends upon the identification of endocardium boundaries as well as the calculation of end-diastolic (ED) and end-systolic (ES) LV volumes. It's critical to segment the LV cavity for precise calculation of EF from echocardiography. Most of the existing echocardiography segmentation approaches either only segment ES and ED frames without leveraging the motion information, or the motion information is only utilized as an auxiliary task. To address the above drawbacks, in this work, we propose a novel echocardiography segmentation method which can effectively utilize the underlying motion information by accurately predicting optical flow (OF) fields. First, we devised a feature extractor shared by the segmentation and the optical flow sub-tasks for efficient information exchange. Then, we proposed a new orientation congruency constraint for the OF estimation sub-task by promoting the congruency of optical flow orientation between successive frames. Finally, we design a motion-enhanced segmentation module for the final segmentation. Experimental results show that the proposed method achieved state-of-the-art performance for EF estimation, with a Pearson correlation coefficient of 0.893 and a Mean Absolute Error of 5.20% when validated with echo sequences of 450 patients.

Automatic view classification of contrast and non-contrast echocardiography.

Background: Contrast and non-contrast echocardiography are crucial for cardiovascular diagnoses and treatments. Correct view classification is a foundational step for the analysis of cardiac structure and function. View classification from all sequences of a patient is laborious and depends heavily on the sonographer's experience. In addition, the intra-view variability and the inter-view similarity increase the difficulty in identifying critical views in contrast and non-contrast echocardiography. This study aims to develop a deep residual convolutional neural network (CNN) to automatically identify multiple views of contrast and non-contrast echocardiography, including parasternal left ventricular short axis, apical two, three, and four-chamber views. Methods: The study retrospectively analyzed a cohort of 855 patients who had undergone left ventricular opacification at the Department of Ultrasound Medicine, Wuhan Union Medical College Hospital from 2013 to 2021, including 70.3% men and 29.7% women aged from 41 to 62 (median age, 53). All datasets were preprocessed to remove sensitive information and 10 frames with equivalent intervals were sampled from each of the original videos. The number of frames in the training, validation, and test datasets were, respectively, 19,370, 2,370, and 2,620 from 9 views, corresponding to 688, 84, and 83 patients. We presented the CNN model to classify echocardiographic views with an initial learning rate of 0.001, and a batch size of 4 for 30 epochs. The learning rate was decayed by a factor of 0.9 per epoch. Results: On the test dataset, the overall classification accuracy is 99.1 and 99.5% for contrast and non-contrast echocardiographic views. The average precision, recall, specificity, and F1 score are 96.9, 96.9, 100, and 96.9% for the 9 echocardiographic views. Conclusions: This study highlights the potential of CNN in the view classification of echocardiograms with and without contrast. It shows promise in improving the workflow of clinical analysis of echocardiography.

Sketch guided and progressive growing GAN for realistic and editable ultrasound image synthesis.

Ultrasound (US) imaging is widely used for anatomical structure inspection in clinical diagnosis. The training of new sonographers and deep learning based algorithms for US image analysis usually requires a large amount of data. However, obtaining and labeling large-scale US imaging data are not easy tasks, especially for diseases with low incidence. Realistic US image synthesis can alleviate this problem to a great extent. In this paper, we propose a generative adversarial network (GAN) based image synthesis framework. Our main contributions include: (1) we present the first work that can synthesize realistic B-mode US images with high-resolution and customized texture editing features; (2) to enhance structural details of generated images, we propose to introduce auxiliary sketch guidance into a conditional GAN. We superpose the edge sketch onto the object mask and use the composite mask as the network input; (3) to generate high-resolution US images, we adopt a progressive training strategy to gradually generate high-resolution images from low-resolution images. In addition, a feature loss is proposed to minimize the difference of high-level features between the generated and real images, which further improves the quality of generated images; (4) the proposed US image synthesis method is quite universal and can also be generalized to the US images of other anatomical structures besides the three ones tested in our study (lung, hip joint, and ovary); (5) extensive experiments on three large US image datasets are conducted to validate our method. Ablation studies, customized texture editing, user studies, and segmentation tests demonstrate promising results of our method in synthesizing realistic US images.

Regional Cardiac Motion Scoring With Multi-Scale Motion-Based Spatial Attention.

Regional cardiac motion scoring aims to classify the motion status of each myocardium segment into one of the four categories (normal, hypokinetic, akinetic, and dyskinetic) from multiple short-axis MR sequences. It is essential for prognosis and early diagnosis for various cardiac diseases. However, the complex motion procedure of the myocardium and the invisible pattern differences pose great challenges, leading to low performance for automatic methods. Most existing works mitigate the task by differentiating the normal motion patterns from the abnormal ones, without fine-grained motion scoring. We propose an effective method for the task of cardiac motion scoring by connecting a bottom-up and another top-down branch with a novel motion-based spatial attention module in multi-scale space. Specifically, we use the convolution blocks for low-level feature extraction that acts as a bottom-up mechanism, and the task of optical flow for explicit motion extraction that acts as a top-down mechanism for high-level allocation of spatial attention. To this end, a newly designed Multi-scale Motion-based Spatial Attention (MMSA) module is used as the pivot connecting the bottom-up part and the top-down part, and adaptively weight the low-level features according to the motion information. Experimental results on a newly constructed dataset of 1440 myocardium segments from 90 subjects demonstrate that the proposed MMSA can accurately analyze the regional myocardium motion, with accuracies of 79.3% for 4-way motion scoring, 89.0% for abnormality detection, and correlation of 0.943 for estimation of motion score index. This work has great potential for practical assessmentof cardiac motion function.

AWSnet: An auto-weighted supervision attention network for myocardial scar and edema segmentation in multi-sequence cardiac magnetic resonance images.

Multi-sequence cardiac magnetic resonance (CMR) provides essential pathology information (scar and edema) to diagnose myocardial infarction. However, automatic pathology segmentation can be challenging due to the difficulty of effectively exploring the underlying information from the multi-sequence CMR data. This paper aims to tackle the scar and edema segmentation from multi-sequence CMR with a novel auto-weighted supervision framework, where the interactions among different supervised layers are explored under a task-specific objective using reinforcement learning. Furthermore, we design a coarse-to-fine framework to boost the small myocardial pathology region segmentation with shape prior knowledge. The coarse segmentation model identifies the left ventricle myocardial structure as a shape prior, while the fine segmentation model integrates a pixel-wise attention strategy with an auto-weighted supervision model to learn and extract salient pathological structures from the multi-sequence CMR data. Extensive experimental results on a publicly available dataset from Myocardial pathology segmentation combining multi-sequence CMR (MyoPS 2020) demonstrate our method can achieve promising performance compared with other state-of-the-art methods. Our method is promising in advancing the myocardial pathology assessment on multi-sequence CMR data. To motivate the community, we have made our code publicly available via https://github.com/soleilssss/AWSnet/tree/master.

Joint Landmark and Structure Learning for Automatic Evaluation of Developmental Dysplasia of the Hip.

The ultrasound (US) screening of the infant hip is vital for the early diagnosis of developmental dysplasia of the hip (DDH). The US diagnosis of DDH refers to measuring alpha and beta angles that quantify hip joint development. These two angles are calculated from key anatomical landmarks and structures of the hip. However, this measurement process is not trivial for sonographers and usually requires a thorough understanding of complex anatomical structures. In this study, we propose a multi-task framework to learn the relationships among landmarks and structures jointly and automatically evaluate DDH. Our multi-task networks are equipped with three novel modules. Firstly, we adopt Mask R-CNN as the basic framework to detect and segment key anatomical structures and add one landmark detection branch to form a new multi-task framework. Secondly, we propose a novel shape similarity loss to refine the incomplete anatomical structure prediction robustly and accurately. Thirdly, we further incorporate the landmark-structure consistent prior to ensure the consistency of the bony rim estimated from the segmented structure and the detected landmark. In our experiments, 1231 US images of the infant hip from 632 patients are collected, of which 247 images from 126 patients are tested. The average errors in alpha and beta angles are 2.221 ° and 2.899 °. About 93% and 85% estimates of alpha and beta angles have errors less than 5 degrees, respectively. Experimental results demonstrate that the proposed method can accurately and robustly realize the automatic evaluation of DDH, showing great potential for clinical application.

Agent With Warm Start and Adaptive Dynamic Termination for Plane Localization in 3D Ultrasound.

Accurate standard plane (SP) localization is the fundamental step for prenatal ultrasound (US) diagnosis. Typically, dozens of US SPs are collected to determine the clinical diagnosis. 2D US has to perform scanning for each SP, which is time-consuming and operator-dependent. While 3D US containing multiple SPs in one shot has the inherent advantages of less user-dependency and more efficiency. Automatically locating SP in 3D US is very challenging due to the huge search space and large fetal posture variations. Our previous study proposed a deep reinforcement learning (RL) framework with an alignment module and active termination to localize SPs in 3D US automatically. However, termination of agent search in RL is important and affects the practical deployment. In this study, we enhance our previous RL framework with a newly designed adaptive dynamic termination to enable an early stop for the agent searching, saving at most 67% inference time, thus boosting the accuracy and efficiency of the RL framework at the same time. Besides, we validate the effectiveness and generalizability of our algorithm extensively on our in-house multi-organ datasets containing 433 fetal brain volumes, 519 fetal abdomen volumes, and 683 uterus volumes. Our approach achieves localization error of 2.52mm/10.26° , 2.48mm/10.39° , 2.02mm/10.48° , 2.00mm/14.57° , 2.61mm/9.71° , 3.09mm/9.58° , 1.49mm/7.54° for the transcerebellar, transventricular, transthalamic planes in fetal brain, abdominal plane in fetal abdomen, and mid-sagittal, transverse and coronal planes in uterus, respectively. Experimental results show that our method is general and has the potential to improve the efficiency and standardization of US scanning.

Left Ventricle Quantification Challenge: A Comprehensive Comparison and Evaluation of Segmentation and Regression for Mid-Ventricular Short-Axis Cardiac MR Data.

Automatic quantification of the left ventricle (LV) from cardiac magnetic resonance (CMR) images plays an important role in making the diagnosis procedure efficient, reliable, and alleviating the laborious reading work for physicians. Considerable efforts have been devoted to LV quantification using different strategies that include segmentation-based (SG) methods and the recent direct regression (DR) methods. Although both SG and DR methods have obtained great success for the task, a systematic platform to benchmark them remains absent because of differences in label information during model learning. In this paper, we conducted an unbiased evaluation and comparison of cardiac LV quantification methods that were submitted to the Left Ventricle Quantification (LVQuan) challenge, which was held in conjunction with the Statistical Atlases and Computational Modeling of the Heart (STACOM) workshop at the MICCAI 2018. The challenge was targeted at the quantification of 1) areas of LV cavity and myocardium, 2) dimensions of the LV cavity, 3) regional wall thicknesses (RWT), and 4) the cardiac phase, from mid-ventricle short-axis CMR images. First, we constructed a public quantification dataset Cardiac-DIG with ground truth labels for both the myocardium mask and these quantification targets across the entire cardiac cycle. Then, the key techniques employed by each submission were described. Next, quantitative validation of these submissions were conducted with the constructed dataset. The evaluation results revealed that both SG and DR methods can offer good LV quantification performance, even though DR methods do not require densely labeled masks for supervision. Among the 12 submissions, the DR method LDAMT offered the best performance, with a mean estimation error of 301 mm 2 for the two areas, 2.15 mm for the cavity dimensions, 2.03 mm for RWTs, and a 9.5% error rate for the cardiac phase classification. Three of the SG methods also delivered comparable performances. Finally, we discussed the advantages and disadvantages of SG and DR methods, as well as the unsolved problems in automatic cardiac quantification for clinical practice applications.

Modality alignment contrastive learning for severity assessment of COVID-19 from lung ultrasound and clinical information.

The outbreak of COVID-19 around the world has caused great pressure to the health care system, and many efforts have been devoted to artificial intelligence (AI)-based analysis of CT and chest X-ray images to help alleviate the shortage of radiologists and improve the diagnosis efficiency. However, only a few works focus on AI-based lung ultrasound (LUS) analysis in spite of its significant role in COVID-19. In this work, we aim to propose a novel method for severity assessment of COVID-19 patients from LUS and clinical information. Great challenges exist regarding the heterogeneous data, multi-modality information, and highly nonlinear mapping. To overcome these challenges, we first propose a dual-level supervised multiple instance learning module (DSA-MIL) to effectively combine the zone-level representations into patient-level representations. Then a novel modality alignment contrastive learning module (MA-CLR) is presented to combine representations of the two modalities, LUS and clinical information, by matching the two spaces while keeping the discriminative features. To train the nonlinear mapping, a staged representation transfer (SRT) strategy is introduced to maximumly leverage the semantic and discriminative information from the training data. We trained the model with LUS data of 233 patients, and validated it with 80 patients. Our method can effectively combine the two modalities and achieve accuracy of 75.0% for 4-level patient severity assessment, and 87.5% for the binary severe/non-severe identification. Besides, our method also provides interpretation of the severity assessment by grading each of the lung zone (with accuracy of 85.28%) and identifying the pathological patterns of each lung zone. Our method has a great potential in real clinical practice for COVID-19 patients, especially for pregnant women and children, in aspects of progress monitoring, prognosis stratification, and patient management.

Left ventricle quantification with sample-level confidence estimation via Bayesian neural network.

Quantification of cardiac left ventricle has become a hot topic due to its great significance in clinical practice. Many efforts have been devoted to LV quantification and obtained promising performance with the help of various deep neural networks when validated on a group of samples. However, none of them can provide sample-level confidence of the results, i.e., how reliable is the prediction result for one single sample, which would help clinicians make decisions of whether or not to accept the automatic results. In this paper, we achieve this by introducing the uncertainty analysis theory into our LV quantification network. Two types of uncertainty, Model Uncertainty, and Data Uncertainty are analyzed for the quantification performance and contribute to the sample-level confidence. Experiments with data of 145 subjects validate that our method not only improved the quantification performance with an uncertainty-weighted regression loss but also is capable of providing for each sample the confidence level of the estimation results for clinicians' further consideration.

Arterial Spin Labeling Images Synthesis From sMRI Using Unbalanced Deep Discriminant Learning.

Adequate medical images are often indispensable in contemporary deep learning-based medical imaging studies, although the acquisition of certain image modalities may be limited due to several issues including high costs and patients issues. However, thanks to recent advances in deep learning techniques, the above tough problem can be substantially alleviated by medical images synthesis, by which various modalities including T1/T2/DTI MRI images, PET images, cardiac ultrasound images, retinal images, and so on, have already been synthesized. Unfortunately, the arterial spin labeling (ASL) image, which is an important fMRI indicator in dementia diseases diagnosis nowadays, has never been comprehensively investigated for the synthesis purpose yet. In this paper, ASL images have been successfully synthesized from structural magnetic resonance images for the first time. Technically, a novel unbalanced deep discriminant learning-based model equipped with new ResNet sub-structures is proposed to realize the synthesis of ASL images from structural magnetic resonance images. The extensive experiments have been conducted. Comprehensive statistical analyses reveal that: 1) this newly introduced model is capable to synthesize ASL images that are similar towards real ones acquired by actual scanning; 2) synthesized ASL images obtained by the new model have demonstrated outstanding performance when undergoing rigorous tests of region-based and voxel-based corrections of partial volume effects, which are essential in ASL images processing; and 3) it is also promising that the diagnosis performance of dementia diseases can be significantly improved with the help of synthesized ASL images obtained by the new model, based on a multi-modal MRI dataset containing 355 demented patients in this paper.

Full left ventricle quantification via deep multitask relationships learning.

Cardiac left ventricle (LV) quantification is among the most clinically important tasks for identification and diagnosis of cardiac disease. However, it is still a task of great challenge due to the high variability of cardiac structure across subjects and the complexity of temporal dynamics of cardiac sequences. Full quantification, i.e., to simultaneously quantify all LV indices including two areas (cavity and myocardium), six regional wall thicknesses (RWT), three LV dimensions, and one phase (Diastole or Systole), is even more challenging since the ambiguous correlations existing among these indices may impinge upon the convergence and generalization of the learning procedure. In this paper, we propose a deep multitask relationship learning network (DMTRL) for full LV quantification. The proposed DMTRL first obtains expressive and robust cardiac representations with a deep convolution neural network (CNN); then models the temporal dynamics of cardiac sequences effectively with two parallel recurrent neural network (RNN) modules. After that, it estimates the three types of LV indices under a Bayesian framework that is capable of learning multitask relationships automatically, and estimates the cardiac phase with a softmax classifier. The CNN representation, RNN temporal modeling, Bayesian multitask relationship learning, and softmax classifier establish an effective and integrated network which can be learned in an end-to-end manner. The obtained task covariance matrix captures the correlations existing among these indices, therefore leads to accurate estimation of LV indices and cardiac phase. Experiments on MR sequences of 145 subjects show that DMTRL achieves high accurate prediction, with average mean absolute error of 180 mm2, 1.39 mm, 2.51 mm for areas, RWT, dimensions and error rate of 8.2% for the phase classification. This endows our method a great potential in comprehensive clinical assessment of global, regional and dynamic cardiac function.

Direct Multitype Cardiac Indices Estimation via Joint Representation and Regression Learning.

Cardiac indices estimation is of great importance during identification and diagnosis of cardiac disease in clinical routine. However, estimation of multitype cardiac indices with consistently reliable and high accuracy is still a great challenge due to the high variability of cardiac structures and the complexity of temporal dynamics in cardiac MR sequences. While efforts have been devoted into cardiac volumes estimation through feature engineering followed by a independent regression model, these methods suffer from the vulnerable feature representation and incompatible regression model. In this paper, we propose a semi-automated method for multitype cardiac indices estimation. After the manual labeling of two landmarks for ROI cropping, an integrated deep neural network Indices-Net is designed to jointly learn the representation and regression models. It comprises two tightly-coupled networks, such as a deep convolution autoencoder for cardiac image representation, and a multiple output convolution neural network for indices regression. Joint learning of the two networks effectively enhances the expressiveness of image representation with respect to cardiac indices, and the compatibility between image representation and indices regression, thus leading to accurate and reliable estimations for all the cardiac indices. When applied with five-fold cross validation on MR images of 145 subjects, Indices-Net achieves consistently low estimation error for LV wall thicknesses (1.44 ± 0.71 mm) and areas of cavity and myocardium (204 ± 133 mm2). It outperforms, with significant error reductions, segmentation method (55.1% and 17.4%), and two-phase direct volume-only methods (12.7% and 14.6%) for wall thicknesses and areas, respectively. These advantages endow the proposed method a great potential in clinical cardiac function assessment.

Blind image quality assessment using joint statistics of gradient magnitude and Laplacian features.

Blind image quality assessment (BIQA) aims to evaluate the perceptual quality of a distorted image without information regarding its reference image. Existing BIQA models usually predict the image quality by analyzing the image statistics in some transformed domain, e.g., in the discrete cosine transform domain or wavelet domain. Though great progress has been made in recent years, BIQA is still a very challenging task due to the lack of a reference image. Considering that image local contrast features convey important structural information that is closely related to image perceptual quality, we propose a novel BIQA model that utilizes the joint statistics of two types of commonly used local contrast features: 1) the gradient magnitude (GM) map and 2) the Laplacian of Gaussian (LOG) response. We employ an adaptive procedure to jointly normalize the GM and LOG features, and show that the joint statistics of normalized GM and LOG features have desirable properties for the BIQA task. The proposed model is extensively evaluated on three large-scale benchmark databases, and shown to deliver highly competitive performance with state-of-the-art BIQA models, as well as with some well-known full reference image quality assessment models.