Segmentation Of Left Ventricle Research Articles

The value of 3D-speckle tracking longitudinal strain for the assessment of left ventricular function recovery in ischemic heart failure patients undergoing surgical remodeling:the RECOVERY-IN study

Abstract Background Three-dimensional (3D) speckle-tracking echocardiography is largely employed to evaluate left ventricle (LV) morphology and function. Purpose To investigate LV function before and after surgical ventricular reconstruction (SVR) through the analysis of global (GLS) and segmental (SLS) longitudinal strain, and the derived mechanical dispersion (MD). Methods Twenty patients eligible for SVR, with previous LV remodelling and ischemic heart failure (HF), received 3D echocardiographic evaluation before SVR and at 6-months follow-up; 15 normal controls, matched by age and BSA, were enrolled. Standard off-line GLS analysis was performed with 4D LV-ANALYSIS©; advanced segmental analysis was accomplished automatically through in-house numerical post-processing. Results Before SVR, GLS deteriorated compared to normal subjects (−6.7% vs. −19.6%, P<0.0001) as confirmed by SLS at each LV segment basal, mid and apical level (P<0.0001); MD was higher than in controls (P<0.001) and markedly increased from basal to apical LV segment. After SVR, GLS significantly improved from −6.7% to −11.3% (P<0.0001). Analysis of variance showed that SLS recovery was higher in the basal region (7.25%) than in both mid (4.06%, P=0.001) and apical (1.92%, P<0.0001) segments, respectively, with adjustment for baseline values. Conclusions After SVR, LV longitudinal strain mostly improves in the basal segments, outlining the role of the remote myocardium in enhancing LV function through an extensive volume reduction; post-surgical MD reduction indicates a more homogeneous myocardial contraction. Heath map of longitudinal strain (%) Funding Acknowledgement Type of funding source: Private hospital(s). Main funding source(s): IRCCS Policlinico San Donato is a clinical research hospital partially funded by the Italian Ministry of Health.

Predictors of left ventricular remodelling in patients after active myocarditis

Abstract Background Myocarditis may be a challenging diagnosis because of diversity of clinical presentations. Thus, the clinical course and ventricular remodelling in this disease seems to be unpredictable. Aims To assess predictors of left ventricular remodelling in patients after active myocarditis. Methods Database from a high-volume, tertiary cardiology center was analysed to identify patients with active myocarditis, based on clinical presentation and ≥1 diagnostic criteria from different categories (including electrocardiography/ holter, elevated troponin T/I levels, functional or structural abnormalities on cardiac imaging or tissue characterization by cardiac magnetic resonance) between 2016 and 2019. Left ventricle global longitudinal strain (GLS), mechanical dispersion (standard deviation of time to peak longitudinal strain derived from all left ventricle segments in 3 apical views) were calculated. Response (left ventricular positive remodeling measured by transthoracic echocardiography) was defined as end-systolic volume (ESV) reduction ≥15% from a baseline value or end-diastolic volume (EDV) reduction ≥15%. Results 61 consecutive patients were enrolled. The median follow-up was 1.4 years (range: 0.3–4.0). During entire follow-up period mortality rate in patients was 1.6%. Multivariate Cox regression model including significant baseline differences as covariates reported that QRS durations (HR 1.31, 95% CI 1.17–1.57, P=0.049) and mechanical dispersion (HR 1.03, 95% CI 1.01–1.07, P=0.036) were independently associated with left ventricular positive remodelling with ESV reduction. Mechanical dispersion (HR 1.04, 95% CI 1.02–1.06, P=0.040) was independently associated with left ventricular positive remodelling with EDV reduction. Conclusions Mechanical dispersion and QRS duration are independent predictors of left ventricular remodelling in patients after active myocarditis. Funding Acknowledgement Type of funding source: None

The value of regional myocardial function assessment in patients with acute myocarditis at baseline and mid term follow-up

Abstract Background The clinical course and ventricular remodeling in inflammatory myocardial disease could be unpredictable. No single functional parameter has been confirmed as a powerful predictor of clinical course and functional recovery assessment in patients with acute inflammatory myocardial disease. Purpose The aim of the study was to assess the mechanical properties of the myocardium in patients with active myocarditis at baseline and follow-up. Methods Database from a high volume, tertiary cardiology center was analysed to identify patients with active myocarditis, based on clinical presentation and ≥1 diagnostic criteria from different categories (including electrocardiography/holter, elevated troponin T/I levels, functional or structural abnormalities on cardiac imaging or tissue characterization by cardiac magnetic resonance) between 2016 and 2019. Conventional and speckle tracking echocardiography including global longitudinal strain (GLS) mechanical dispersion (MD) was completed at baseline and at 17±13 months follow-up. MD was calculated as a standard deviation of time to peak longitudinal strain derived from all left ventricle segments in 3 apical views. Results 61 consecutive patients [50 M, 11F, end-diastolic volume 212±84 ml, end-systolic volume 130±90ml, ejection fraction (EF) 42±16%] were enrolled. During the entire follow-up 1 patient died at early observation. Implantable cardioverter-defibrillator was implanted in 5 patients (primary prevention 4, secondary 1), cardiac resynchronization therapy pacemaker in 1 patient. Despite of significant global improvement (EF 42±16% vs 52±10%, p<0.001) the limited regional improvement was noticed (GLS 14±6% vs 15±4%, p = NS; MD 47±18 ms vs 45±20 ms, p=NS) in all patients at 17±13 months follow-up. There was a strong negative association between GLS and MD at baseline (Figure 1), and slightly weaker at follow-up (R=0.47, Pearson's correlation). Moreover, the GLS correlated well the change of MD in each individual patient. Conclusions Mechanical dispersion and global longitudinal strain may serve as an additional markers of myocardial damage and potential predictive markers in non ischemic cardiomyopathy patients with proven inflammatory origin. Funding Acknowledgement Type of funding source: None

Variation of cardiac magnetic resonance radiomics features by age and sex in healthy participants from the UK Biobank

Abstract Introduction Cardiac magnetic resonance (CMR) radiomics use voxel-level data to derive quantitative indices of myocardial tissue texture, which may provide complementary risk information to traditional CMR measures. Purpose In this first stage of our work, establishing the performance characteristics of CMR radiomics in relation to disease outcomes, we aimed to elucidate differences in radiomic features by sex and age in apparently healthy adults. Methods We defined a healthy cohort from the first 5,065 individuals completing the UK Biobank Imaging Enhancement, limiting to white Caucasian ethnicity, and excluding those with major co-morbidities, or cardiovascular risk factors/symptoms. We created evenly distributed age groups: 45–54 years, 55–64 years, 65–74 years. Radiomics features were extracted from left ventricle segmentations, with normalisation to body surface area. We compared mean values of individual features between the sexes, stratified by age and separately between the oldest and youngest age groups for each sex. Results We studied 657 (309 men, 358 women) healthy individuals. There were significant differences between radiomics features of men and women. Different features appeared more important at different age groups. For instance, in the youngest age group “end-systolic coarseness” showed greatest difference between men and women, whilst “end-diastolic run percentage” and “end-diastolic high grey level emphasis” showed most variation in the oldest and middle age groups. In the oldest age groups, differences between men and women were most predominant in the texture features, whilst in the younger groups a mixture of shape and texture differences were observed. We demonstrate significant variation between radiomics features by age, these differences are exclusively in texture features with different features implicated in men and women (“end-diastolic mean intensity” in women, “end-systolic sum entropy in men”). Conclusions There are significant age and sex differences in CMR radiomics features of apparently healthy adults, demonstrating alterations in myocardial architecture not appreciated by conventional indices. In younger ages, shape and texture differences are observed, whilst in older ages texture differences dominate. Furthermore, texture features are the most different features between the youngest and oldest hearts. We provide proof-of-concept data indicating CMR radiomics has discriminatory value with regard to two characteristics strongly linked to cardiovascular outcomes. We will next elucidate relationships between CMR radiomics, cardiac risk factors, and clinical outcomes, establishing predictive value incremental to existing measures. Funding Acknowledgement Type of funding source: Other. Main funding source(s): European Union's Horizon 2020 research and innovation programme (825903),British Heart Foundation Clinical Research Training Fellowship (FS/17/81/33318)

Decreased layer specific longitudinal strain in myocardial segments supplied by stenosed LAD coronary artery: analysis for baseline, peak and recovery phase of stress echocardiography

Abstract Background Despite wide utility of global longitudinal strain in the assessment of left ventricular function up to now little is known about layer specific quantification of myocardial function during stress tests and its potential for ischemia detection. Purpose To investigate diagnostic utility of layer specific LS during stress echocardiography using marker segments in patients with established Left Anterior Descending (LAD) coronary artery status. Methods We retrospectively analyzed 63 patients (mean age 62.5±9.9 years, 62% males) with suspected coronary artery disease, who had performed dobutamine stress echocardiography (DSE). Layer specific speckle tracking LS analyses were performed during rest, maximal stress and recovery. As a marker for LAD (segment most probably supplied by this artery) we used mid-anterior left ventricle segments. LAD status were established with coronary angiography or angioCT and consider as significant if ≥70% stenosis was present. Results In our study group 14 patients had significantly stenosed LAD (group 1) and 24 had normal coronaries (group 2). We detected significantly lowered absolute values of strain in endocardial and mid-myocardial layers of segments supplied by stenosed LAD at rest for endocardium −8.6% (group 1) vs −13.4% (group 2); p=0.045; and mid-myocardium −7.3% (group 1) vs −11.2% (group 2); p=0.044. Moreover, we observed similar trend at recovery, however without reaching statistical significance. Conclusions Longitudinal strain analysis focused on segments supplied by assessed coronary artery my provide quantitative data confirming ischemia in the settings of rest and stress echocardiography. Our pilot study suggest that for inotropic agents like dobutamine the potential ischemic dysfunction may be however masked at peak stage of the test. Funding Acknowledgement Type of funding source: None

Echocardiographic image segmentation using deep Res-U network

Cardiac function assessment using echocardiography is a crucial step in daily cardiology. However, cardiac boundary segmentation and in particular, ventricle segmentation is a challenging procedure due to shadows and speckle noise. Manual segmentation of the cardiac boundary is a time-consuming process which rules out conventional segmentation for many situations such as emergency cases and image-guided robotic interventions. Therefore, providing an efficient and robust autonomous segmentation method is crucial for such applications. In this paper, a fast and fully automatic deep learning framework for left ventricle segmentation is proposed. This model couples the advantages of ResNet and U-Net to provide reliable segmentation results. We propose a new encoder in the U-Net, defined as ResU which is a modified version of ResNet-50 and has a superiority over ResNet in data denoising. We trained this model on the dataset CAMUS (Cardiac Acquisitions for Multi-structure Ultrasound Segmentation) which is a large, publicly available and fully annotated dataset for 2D echocardiographic assessment. It is shown that this model outperforms other state-of-the-art methods in terms of accuracy with a Dice metric of 0.97±0.01.

Reduced Left Ventricular Myocardial Blood Flow After Radiation Therapy For Thoracic Esophagus Cancer: A Study Using Stress Dynamic Perfusion CT

Radiation therapy (RT) has greatly contributed to improving the prognosis of cancer in the thoracic region. As treatment outcomes improve, myocardial disorders induced by RT have become a serious problem. To investigate myocardial disorder after RT for esophageal cancer using myocardial stress dynamic computed tomography perfusion (CTP) before and after RT. Fifteen patients with esophageal cancer scheduled for RT were prospectively enrolled between January 2017 and January 2019. CTP was performed before and 3 months after RT to investigate the relationship between changes in hyperemic myocardial blood flow (hMBF) and the radiation dose for the left ventricle (LV). Fifteen segments of LV myocardium were classified into 3 groups according to radiation dose: <20 Gy (Group A), 20–40 Gy (Group B) and >40 Gy (Group C), by fusing CT angiography (CTA) onto radiation planning CT images. Twelve of 15 patients completed CTP before and at a median of 2.8±1.34 months after RT. A total of 180 LV segments (Group A, n = 122; B, n = 39; and C, n = 19) was analyzed. Groups B (pre-RT 147.5 ± 27.7 mL/min/100g vs post-RT 135.5 ± 25.4 mL/min/100g, p = 0.035) and C(pre-RT 146.3 ± 27.0 mL/min/100g vs post-RT 126.0 ± 16.0 mL/min/100g, p = 0.022) showed significant reduction in hMBF, while there was no significant difference in hMBF for Group A between pre- and post-RT (145.3 ± 26.3 mL/min/100g vs 146.4 ± 19.3 mL/min/100g, p = 0.729). There was a significant difference in ΔhMBF between Groups A and B (1 ± 27.5 mL/min/100g vs -12 ± 33.9 mL/min/100g, p = 0.014). Significant negative correlation was found between the LV radiation dose and change in hMBF (r = -0.263, p = 0.004). Significant reduction in hMBF was observed with CTP taken approximately 3 months after RT in the LV segments for which radiation dose ≥20 Gy was administered. The alteration in hMBF was related to LV radiation dose. These findings should increase our knowledge of ionizing radiation injury to the heart, and provide important insights for RT planning optimization to prevent or mitigate the adverse effects.

Diagnostic performance of an artificial intelligence-driven cardiac-structured reporting system for myocardial perfusion SPECT imaging

ObjectivesTo describe and validate an artificial intelligence (AI)-driven structured reporting system by direct comparison of automatically generated reports to results from actual clinical reports generated by nuclear cardiology experts. BackgroundQuantitative parameters extracted from myocardial perfusion imaging (MPI) studies are used by our AI reporting system to generate automatically a guideline-compliant structured report (sR). MethodA new nonparametric approach generates distribution functions of rest and stress, perfusion, and thickening, for each of 17 left ventricle segments that are then transformed to certainty factors (CFs) that a segment is hypoperfused, ischemic. These CFs are then input to our set of heuristic rules used to reach diagnostic findings and impressions propagated into a sR referred as an AI-driven structured report (AIsR).The diagnostic accuracy of the AIsR for detecting coronary artery disease (CAD) and ischemia was tested in 1,000 patients who had undergone rest/stress SPECT MPI. ResultsAt the high-specificity (SP) level, in a subset of 100 patients, there were no statistical differences in the agreements between the AIsr, and nine experts’ impressions of CAD (P = .33) or ischemia (P = .37). This high-SP level also yielded the highest accuracy across global and regional results in the 1,000 patients. These accuracies were statistically significantly better than the other two levels [sensitivity (SN)/SP tradeoff, high SN] across all comparisons. ConclusionsThis AI reporting system automatically generates a structured natural language report with a diagnostic performance comparable to those of experts.

Dose-sparing effect of deep inspiration breath hold technique on coronary artery and left ventricle segments in treatment of breast cancer

Background and purposeThe risk of radiation-induced cardiac injury remains a challenging problem in the treatment of breast cancer. Certain cardiac structures receive higher doses than others, which results in variable frequencies of radiation-induced injuries across these structures. Radiation dose can be reduced using the deep inspiration breath hold (DIBH) technique. We aimed to investigate the dose reductions from DIBH in individual cardiac segments. Materials and methodsA dosimetric analysis was performed on left-sided breast cancer patients who underwent breast-conserving surgery and whole breast irradiation. Radiation doses to the cardiac structures were compared between the DIBH and free-breathing (FB) techniques and the dose reductions with DIBH were correlated to the lung expansion. ResultsFor the 75 patients included in our study, DIBH effectively reduced doses to the heart, left lung, left anterior descending coronary artery (LAD) and left ventricle (LV), but the degree of dose reductions was variable across different structures. The absolute dose reductions were greatest in the distal LAD (14.4 Gy) and apical LV (12.1 Gy) segments, compared with the other LAD (middle 9.7 Gy, proximal 1.6 Gy) and LV (anterior 5.3 Gy, lateral 2.9 Gy, septal 2.0 Gy, inferior 0.2 Gy) segments. Left lung expansion was significantly correlated with the dose reductions in the LAD (Spearman’s rank correlation coefficient, ρ, 0.304) and LV (ρ, 0.420) segments. ConclusionsOur study demonstrates the dose-sparing effects of DIBH in various cardiac structures, especially the distal LAD and apical LV segments. The large dose reductions seen in the distal LAD and apical LV segments could potentially translate into clinical benefit of reduced cardiac toxicity, as these structures have been previously shown to receive the highest doses and are associated with radiation-induced injury.

Automatic cardiac MRI segmentation and permutation-invariant pathology classification using deep neural networks and point clouds

Segmentation of cardiac MRI images plays a key role in clinical diagnosis. In the traditional diagnostic process, clinical experts manually segment left ventricle (LV), right ventricle (RV) and myocardium (Myo) to get the guideline for cardiopathy diagnosis. However, manual segmentation is time-consuming and labor-intensive. In this paper, we propose automatic cardiac MRI segmentation and cardiopathy classification based on deep neural networks and point clouds. The cardiac MRI segmentation consists of two steps: (i) We use a detector based on you only look once (YOLO) to obtain region of interest (ROI) from the sequential diastolic and systolic MRI. (ii) We obtain the segmentation masks from the ROI automatically by a fully convolutional neural network (FCN). Subsequently, we reconstruct 3D surfaces by a simple linear interpolation method, then randomly sample uniform 3D point clouds from the 3D surfaces. From the cardiac point clouds, we perform cardiopathy classification using a cardiopathy diagnosis network (CDN). Experimental results show that the proposed method successfully segments LV, RV, and Myo from cardiac MRI images and achieves comparable results against several existing ones. Moreover, the CDN successfully classifies heart diseases based on point clouds and achieves 92% accuracy on the testing dataset.

Fully automatic segmentation of right and left ventricle on short-axis cardiac MRI images

Cardiac magnetic resonance imaging (CMR) is a widely used non-invasive imaging modality for evaluating cardiovascular diseases. CMR is the gold standard method for left and right ventricular functional assessment due to its ability to characterize myocardial structure and function and low intra- and inter-observer variability. However the post-processing segmentation during the functional evaluation is time-consuming and challenging. A fully automated segmentation method can assist the experts; therefore, they can do more efficient work. In this paper, a regression-based fully automated method is presented for the right- and left ventricle segmentation. For training and evaluation, our dataset contained MRI short-axis scans of 5570 patients, who underwent CMR examinations at Heart and Vascular Center, Semmelweis University Budapest. Our approach is novel and after training the state-of-the-art algorithm on our dataset, our algorithm proved to be superior on both of the ventricles. The evaluation metrics were the Dice index, Hausdorff distance and volume related parameters. We have achieved average Dice index for the left endocardium: 0.927, left epicardium: 0.940 and right endocardium: 0.873 on our dataset. We have also compared the performance of the algorithm to the human-level segmentation on both ventricles and it is similar to experienced readers for the left, and comparable for the right ventricle. We also evaluated the proposed algorithm on the ACDC dataset, which is publicly available, with and without transfer learning. The results on ACDC were also satisfying and similar to human observers. Our method is lightweight, fast to train and does not require more than 2 GB GPU memory for execution and training.

Quantification of Myocardial Dosimetry and Glucose Metabolism Using a 17-Segment Model of the Left Ventricle in Esophageal Cancer Patients Receiving Radiotherapy.

ObjectivePrevious studies have shown that increased cardiac uptake of 18F-fluorodeoxyglucose (FDG) on positron emission tomography (PET) may be an indicator of myocardial injury after radiotherapy (RT). The primary objective of this study was to quantify cardiac subvolume dosimetry and 18F-FDG uptake on oncologic PET using a 17-segment model of the left ventricle (LV) and to identify dose limits related to changes in cardiac 18F-FDG uptake after RT.MethodsTwenty-four esophageal cancer (EC) patients who underwent consecutive oncologic 18F-FDG PET/CT scans at baseline and post-RT were enrolled in this study. The radiation dose and the 18F-FDG uptake were quantitatively analyzed based on a 17-segment model. The 18F-FDG uptake and doses to the basal, middle and apical regions, and the changes in the 18F-FDG uptake for different dose ranges were analyzed.ResultsA heterogeneous dose distribution was observed, and the basal region received a higher median mean dose (18.36 Gy) than the middle and apical regions (5.30 and 2.21 Gy, respectively). Segments 1, 2, 3, and 4 received the highest doses, all of which were greater than 10 Gy. Three patterns were observed for the myocardial 18F-FDG uptake in relation to the radiation dose before and after RT: an increase (5 patients), a decrease (13 patients), and no change (6 patients). In a pairing analysis, the 18F-FDG uptake after RT decreased by 28.93 and 12.12% in the low-dose segments (0–10 Gy and 10–20 Gy, respectively) and increased by 7.24% in the high-dose segments (20–30 Gy).ConclusionThe RT dose varies substantially within LV segments in patients receiving thoracic EC RT. Increased 18F-FDG uptake in the myocardium after RT was observed for doses above 20 Gy.

Deep RetinaNet for Dynamic Left Ventricle Detection in Multiview Echocardiography Classification

Background. Currently, echocardiography has become an essential technology for the diagnosis of cardiovascular diseases. Accurate classification of apical two-chamber (A2C), apical three-chamber (A3C), and apical four-chamber (A4C) views and the precise detection of the left ventricle can significantly reduce the workload of clinicians and improve the reproducibility of left ventricle segmentation. In addition, left ventricle detection is significant for the three-dimensional reconstruction of the heart chambers. Method. RetinaNet is a one-stage object detection algorithm that can achieve high accuracy and efficiency at the same time. RetinaNet is mainly composed of the residual network (ResNet), the feature pyramid network (FPN), and two fully convolutional networks (FCNs); one FCN is for the classification task, and the other is for the border regression task. Results. In this paper, we use the classification subnetwork to classify A2C, A3C, and A4C images and use the regression subnetworks to detect the left ventricle simultaneously. We display not only the position of the left ventricle on the test image but also the view category on the image, which will facilitate the diagnosis. We used the mean intersection-over-union (mIOU) as an index to measure the performance of left ventricle detection and the accuracy as an index to measure the effect of the classification of the three different views. Our study shows that both classification and detection effects are noteworthy. The classification accuracy rates of A2C, A3C, and A4C are 1.000, 0.935, and 0.989, respectively. The mIOU values of A2C, A3C, and A4C are 0.858, 0.794, and 0.838, respectively.

Automatic segmentation of left ventricle using parallel end–end deep convolutional neural networks framework

Under the background of high incidence and mortality of cardiovascular diseases, the accurate and automatic left ventricle (LV) segmentation method is of essential importance for the diagnosis of the cardiovascular system. However, fully automatic LV segmentation remains challenging due to the complex structure of cardiac magnetic resonance image (MRI) and the morphological changes of LV caused by various cardiovascular diseases. In this paper, we propose a novel parallel end-to-end convolutional neural network (CNN) for LV segmentation. Our network consists of two interactive subnetworks which utilize essentially identical but formally different labels in the hope that they can learn segmentation from different perspectives. The two subnetworks take the same cardiac MRI as input and output a pair of segmentation maps in different forms. After averaging the two segmentation maps obtained from the two subnetworks, we get the final contours of the endocardium (endo) and epicardium (epi) simultaneously. The proposed method is evaluated on the dataset provided by the Left Ventricle Full Quantification Challenge of MICCAI 2019. The average Dice scores on epi, endo, and myocardium (myo) reach 0.961, 0.949, and 0.867 respectively which outperform the other methods. The experimental results show that our method has the potential for clinical application.

MRI Manufacturer Shift and Adaptation: Increasing the Generalizability of Deep Learning Segmentation for MR Images Acquired with Different Scanners.

To quantitatively evaluate the generalizability of a deep learning segmentation tool to MRI data from scanners of different MRI manufacturers and to improve the cross-manufacturer performance by using a manufacturer-adaptation strategy. This retrospective study included 150 cine MRI datasets from three MRI manufacturers, acquired between 2017 and 2018 (n = 50 for manufacturer 1, manufacturer 2, and manufacturer 3). Three convolutional neural networks (CNNs) were trained to segment the left ventricle (LV), using datasets exclusively from images from a single manufacturer. A generative adversarial network (GAN) was trained to adapt the input image before segmentation. The LV segmentation performance, end-diastolic volume (EDV), end-systolic volume (ESV), LV mass, and LV ejection fraction (LVEF) were evaluated before and after manufacturer adaptation. Paired Wilcoxon signed rank tests were performed. The segmentation CNNs exhibited a significant performance drop when applied to datasets from different manufacturers (Dice reduced from 89.7% ± 2.3 [standard deviation] to 68.7% ± 10.8, P < .05, from 90.6% ± 2.1 to 59.5% ± 13.3, P < .05, from 89.2% ± 2.3 to 64.1% ± 12.0, P < .05, for manufacturer 1, 2, and 3, respectively). After manufacturer adaptation, the segmentation performance was significantly improved (from 68.7% ± 10.8 to 84.3% ± 6.2, P < .05, from 72.4% ± 10.2 to 85.7% ± 6.5, P < .05, for manufacturer 2 and 3, respectively). Quantitative LV function parameters were also significantly improved. For LVEF, the manufacturer adaptation increased the Pearson correlation from 0.005 to 0.89 for manufacturer 2 and from 0.77 to 0.94 for manufacturer 3. A segmentation CNN well trained on datasets from one MRI manufacturer may not generalize well to datasets from other manufacturers. The proposed manufacturer adaptation can largely improve the generalizability of a deep learning segmentation tool without additional annotation.Supplemental material is available for this article.© RSNA, 2020.

ResDUnet: Residual Dilated UNet for Left Ventricle Segmentation from Echocardiographic Images.

Echocardiography is the modality of choice for the assessment of left ventricle function. Left ventricle is responsible for pumping blood rich in oxygen to all body parts. Segmentation of this chamber from echocardiographic images is a challenging task, due to the ambiguous boundary and inhomogeneous intensity distribution. In this paper we propose a novel deep learning model named ResDUnet. The model is based on U-net incorporated with dilated convolution, where residual blocks are employed instead of the basic U-net units to ease the training process. Each block is enriched with squeeze and excitation unit for channel-wise attention and adaptive feature re-calibration. To tackle the problem of left ventricle shape and size variability, we chose to enrich the process of feature concatenation in U-net by integrating feature maps generated by cascaded dilation. Cascaded dilation broadens the receptive field size in comparison with traditional convolution, which allows the generation of multi-scale information which in turn results in a more robust segmentation. Performance measures were evaluated on a publicly available dataset of 500 patients with large variability in terms of quality and patients pathology. The proposed model shows a dice similarity increase of 8.4% when compared to deeplabv3 and 1.2% when compared to the basic U-net architecture. Experimental results demonstrate the potential use in clinical domain.

Improving the generalization of deep learning methods to segment the left ventricle in short axis MR images.

Cardiovascular disease is one of the major health problems worldwide. In clinical practice, cardiac magnetic resonance imaging (CMR) is considered the gold-standard imaging modality for the evaluation of the function and structure of the left ventricle (LV). More recently, deep learning methods have been used to segment LV with impressive results. On the other hand, this kind of approach is prone to overfit the training data, and it does not generalize well between different data acquisition centers, thus creating constraints to the use in daily routines. In this paper, we explore methods to improve the generalization in the segmentation performed by a convolutional neural network. We applied a U-net based architecture and compared two different pre-processing methods to improve uniformity in the image contrast between five cross-dataset training and testing. Overall, we were able to perform the segmentation of the left ventricle using multiple cross-dataset combinations of train and test, with a mean endocardium dice score of 0.82.Clinical Relevance- This work improves the result between the cross-dataset evaluation of the left ventricle segmentation, reducing the constraints for daily clinical adoption of a fully-automatic segmentation method.

Automatic left ventricle segmentation in short-axis MRI using deep convolutional neural networks and central-line guided level set approach

In the clinical diagnosis of cardiovascular diseases, left ventricle (LV) segmentation in cardiac magnetic resonance images (MRI) is an indispensable procedure for doctors. To reduce the time needed for diagnosis, we develop an automatic LV segmentation method by integrating the convolutional neural network (CNN) with the level set approach. Firstly, a CNN based myocardial central-line detection algorithm was proposed to replace the manual initialization process for traditional level set approaches. Secondly, we present a novel central-line guided level set approach (CGLS) for delineating the myocardium region. In particular, we incorporate the myocardial central-line into the level set energy formulation as a constraint term. It plays two important roles in the iterative process: restricting the zero-level contour to stay around the myocardial central-line and preserving the anatomical geometry of myocardium segmentation result. In experiments, our method yields results as below: (1) 1.74 mm and 2.06 mm in terms of epicardium and endocardium perpendicular distance on MICCAI 2009 dataset, (2) 0.955 and 0.853 in terms of LV and myocardium Dice metric at the end-diastole on ACDC MICCAI 2017 dataset. The experimental data demonstrate that our method outperforms some state-of-the-art methods and achieves a good agreement with the manual segmentation results.

Automated mitral valve vortex ring extraction from 4D-flow MRI.

PurposeTo present and validate a method for automated extraction and analysis of the temporal evolution of the mitral valve (MV) vortex ring from MR 4D‐flow data.MethodsThe proposed algorithm uses the divergence‐free part of the velocity vector field for Q criterion‐based identification and tracking of MV vortex ring core and region within the left ventricle (LV). The 4D‐flow data of 20 subjects (10 healthy controls, 10 patients with ischemic heart disease) were used to validate the algorithm against visual analysis as well as to assess the method’s sensitivity to manual LV segmentation. Quantitative MV vortex ring parameters were analyzed with respect to both their differences between healthy subjects and patients and their correlation with transmitral peak velocities.ResultsThe algorithm successfully extracted MV vortex rings throughout the entire cardiac cycle, which agreed substantially with visual analysis (Cohen’s kappa = 0.77). Furthermore, vortex cores and regions were robustly detected even if a static end‐diastolic LV segmentation mask was applied to all frames (Dice coefficients 0.82 ± 0.08 and 0.94 ± 0.02 for core and region, respectively). Early diastolic MV vortex ring vorticity, kinetic energy and circularity index differed significantly between healthy controls and patients. In contrast to vortex shape parameters, vorticity and kinetic energy correlated strongly with transmitral peak velocities.ConclusionAn automated method for temporal MV vortex ring extraction demonstrating robustness with respect to LV segmentation strategies is introduced. Quantitative vortex parameter analysis indicates importance of the MV vortex ring for LV diastolic (dys)function.

Left ventricle segmentation using a Bayesian approach with distance dependent shape priors

We propose a method for segmentation of the left ventricle in magnetic resonance cardiac images. The framework consists of an initial Bayesian segmentation of the central slice of the volume. This segmentation is used to locate a shape prior for the LV myocardial tissue. This shape prior is determined using the fact that the myocardium is approximately annular as seen in the short-axis. Then a second Bayesian segmentation is performed to obtain the final result. This procedure is repeated for the rest of the slices. An extrapolation of the area of the LV is used to determine a stopping criterion. The method was evaluated on the databases of the Cardiac Atlas project. Our results demonstrate a suitable accuracy for myocardial segmentation (≈0.8 Dice’s coefficient). For the endocardium and the epicardium the Dice’s coefficients are 0.94 and 0.9 respectively. The accuracy was also evaluated in terms of the Hausdorff distance and the average distance. For the myocardium we obtain 8 mm and 2 mm respectively. Our results demonstrate the capability and merits of the proposed method to estimate the structure of the LV. The method requires minimal user input and generates results with quality comparable to more complex approaches. This paper suggests a new efficient approach for automatic LV quantification based on a Bayesian technique with shape priors with errors comparable to state-of-the-art techniques.