Heart Segmentation Research Articles

Deep Learning-Based Computer-Aided Fetal Echocardiography: Application to Heart Standard View Segmentation for Congenital Heart Defects Detection.

Accurate segmentation of fetal heart in echocardiography images is essential for detecting the structural abnormalities such as congenital heart defects (CHDs). Due to the wide variations attributed to different factors, such as maternal obesity, abdominal scars, amniotic fluid volume, and great vessel connections, this process is still a challenging problem. CHDs detection with expertise in general are substandard; the accuracy of measurements remains highly dependent on humans’ training, skills, and experience. To make such a process automatic, this study proposes deep learning-based computer-aided fetal heart echocardiography examinations with an instance segmentation approach, which inherently segments the four standard heart views and detects the defect simultaneously. We conducted several experiments with 1149 fetal heart images for predicting 24 objects, including four shapes of fetal heart standard views, 17 objects of heart-chambers in each view, and three cases of congenital heart defect. The result showed that the proposed model performed satisfactory performance for standard views segmentation, with a 79.97% intersection over union and 89.70% Dice coefficient similarity. It also performed well in the CHDs detection, with mean average precision around 98.30% for intra-patient variation and 82.42% for inter-patient variation. We believe that automatic segmentation and detection techniques could make an important contribution toward improving congenital heart disease diagnosis rates.

Light-Weight Localization and Scale-Independent Multi-gate UNET Segmentation of Left and Right Ventricles in MRI Images.

Heart segmentation in cardiac magnetic resonance images is heavily used during the assessment of left ventricle global function. Automation of the segmentation is crucial to standardize the analysis. This study aims at developing a CNN-based framework to aid the clinical measurements of the left ventricle and right ventricle in cardiac magnetic resonance images. We propose a fully automated framework for localization and segmentation of the left ventricle and right ventricle in both short- and long-axis views from cardiac magnetic resonance images. The localization module utilizes a light-weight model that detects the region of interest and feeds it to the segmentation model. Also, we propose the Multi-Gate block as an extension to the UNet to boost the segmentation performance by aggregating multi-scale features. Comparison between our proposed method and the baseline UNet was performed to show the gain in the overall performance. The reliability of the model was assessed by testing the method against cardiac magnetic resonance images with different levels of noise and deformations. Heart localization accuracy was 0.59 and 1.75 pixels in both short- and long-axis views respectively. Left and right ventricle blood-pool segmentation Dice was (0.93, 0.90) in end-systole and (0.97, 0.95) in end-diastole. The left ventricle myocardium was segmented accurately with Dice of 0.91 and 0.90 in end-systole and end-diastole respectively. Left ventricle ejection fraction was found to be highly correlated with the gold standard with r = 0.987. Moreover, the proposed pipeline is fast, achieving 0.002 sec per image on average. Adding the Multi-Gate Dilated Inception Block has boosted the performance of UNet architecture and has shown generalization ability when tested on noisy and deformed cardiac magnetic resonance images. The proposed method has proven its wide applicability and reliability for heart detection when tested on different datasets.

Machine Learning Based Analysis of Heart and Cardiac Substructures for Radiation-Induced Cardiac Toxicity From Breast Cancer Radiotherapy

Radiation-induced cardiac toxicity is an importantissue of breast RT. Although a relationship between mean heart dose and acute coronary event (ACE) risk was reported in previous study, the specific relationships between doses to cardiac structures or their sub-regions and subsequent toxicity have not been well defined. In this study, we studied the effect of radiation on specific heart segments on subsequent cardiac toxicity.From the 1294 patients with breast cancer where previously demonstrated a relationship between mean heart dose and ACE, 21 patients who developed HF following breast RT and 39 with ACE were identified after an independent review by cardiologists. The study included 21 and 39 controls for respective toxicity who were matched according to age, comorbidities, and physical activity. For each patient who received radiotherapy, substructures were manually drawn, and then dose-distribution parameters of heart and substructures (Mean, Maximum (Max) dose, Relative volume at least given dose from 5-50 Gy in 5 Gy increments) were collected from CT planning data. To specify most critical regions for developing ACE, analysis using High-dimensional Cox regression model (SNet) was performed to reveal the effect of radiation on specific heart segments on subsequent cardiac toxicity.In HF (n = 42) and ACE (n = 78) cohorts, characteristics between case and control were well matched. Mean time to develop heart failure, acute coronary events were 40.0 months and 42.2 months, respectively. In HF cohort, there was no significant difference not only in the mean heart dose (3.7 Gy vs 3.1 Gy, P = 0.652), but also in the mean left ventricle (LV) (5.0 Gy vs 4.1 Gy, P = 0.587), mean right ventricle (RV) (4.8 Gy vs 3.6 Gy, P = 0.473). On the other hand, in ACE cohort, there was significant difference in mean dose of heart (2.9 Gy vs 1.3 Gy, P = 0.015), left atrium (LA) (1.0 Gy vs 0.5 Gy, P = 0.008), LV (3.7 Gy vs 1.5 Gy, P = 0.028), left anterior descending artery (LAD) (6.6 Gy vs 2.4 Gy, P = 0.028) and right atrium (RA) (1.5 Gy vs 0.9 Gy, P = 0.025). In univariate analysis, mean dose of heart, LA, LV, LAD and RA was significant factors for acute coronary event free survival. ((Heart; P = 0.006, Hazard ratio (HR) 1.139, 95% Confidence Interval (CI) 1.035-1.253) (LA; P = 0.002, HR 1.735, 95% CI 1.209-2.490), (LV; P = 0.018, HR 1.079, 95% CI 1.012-1.150), (LAD; P = 0.031, HR 1.032, 95% CI 1.002-1.064), (RA; P = 0.008, HR = 1.307, 95% CI 1.068-1.598)). For predicting ACE risk for breast cancer patients, SNet model suggested nomogram based on mean and max LA dose, LAD V10, 15, 20, mean heart dose, mean RA dose and RV V35 in the order of higher predicted point variation. Based on this risk stratification model, event free survival significantly showed difference (P = 0.013).There was significant difference between cardiac substructure dose parameters in ACE cohort, but not in HF cohort. Further analysis using convolution neural network will be performed to specify most critical region to be spared.

Subclinical left ventricle impairment following breast cancer radiotherapy: Is there an association between segmental doses and segmental strain dysfunction?

BackgroundCardiotoxicity following breast cancer radiotherapy (RT) represents one of the most redoubtable toxicities. The Global longitudinal strain measurement (GLS) based on 2D speckle tracking imaging (STI) allows detection of left ventricular (LV) dysfunction at a subclinical stage. The aim of this prospective study was to detect patients at risk of cardiotoxicity using echocardiographic parameters and to determine the association between segmental RT doses and early cardiac toxicity. Material and methodsThe STI was performed prior to RT and at 3, 6 and 12 months after.The association between subclinical LV dysfunction, defined as a reduction of GLS more than 10% from the initial value, radiation doses to different LV segments and non-radiation factors were performed based on multivariate analyses. ResultsFrom June 2017 to August 2018, a total of 103 female patients were included. Sixty patients had left sided RT. Seven patients (7.8%) developed a GSL impairment. The segmental alterations predominated in the anteroseptal and apical LV segments. The mean Dmean in altered segments was significantly higher than in non-altered segments (6.7 ± 8.8Gy-7.8 ± 8.9Gy vs 4.9 ± 7.9–5.4 ± 8.2Gy; p < 0.05). Age > 55 years and obesity were important confounding factors that should be considered during radiotherapy planning. ConclusionThe results of our study show that radiation dose is correlated with the subclinical LV segments' alteration. Global heart delineation seems to be insufficient during the breast radiotherapy planning. Segmental delineation of the LV may be an interesting alternative to limit segmental doses and to reduce the risk of subclinical alterations. A mean dose of 5Gy could be proposed in exposed heart segment.

Automated quantification of epicardial adipose tissue on CCTA via deep-learning detection of the pericardium: clinical implications

Abstract Background Epicardial adipose tissue (EAT) is a visceral fat deposit within the pericardial sac which surrounds the heart myocardium and coronary arteries. EAT volume has been demonstrated to be strongly associated with the development and prognosis of cardiovascular diseases, but its measurement is subjective and challenging in practice. Purpose To develop a deep-learning approach for automated segmentation of EAT from routine CCTA scans, that could assist clinical interpretation of CCTA. Methods A deep-learning method using a 3D Residual-U-Net neural network architecture for 3D volumetric segmentation of CCTA data was created. The network was trained on a diverse sample of 1900 CCTAs, each manually segmented by a single expert, drawn from the UK sites of the Oxford Risk Factors And Non-invasive imaging (ORFAN) Study. Three iterations of feedback learning were used to fine tune the algorithm for the segmentation of the whole heart within the bounds of the pericardium. In each iteration, the machine analysed sets of 100–250 unannotated CCTAs unseen by the machine which were then corrected by experts. EAT volumes were calculated by automated thresholding of adipose tissue (−190HU through −30HU) from within the bound of the pericardial segment (Figure 1). The network was then applied to 817 unseen CCTAs from US sites of the ORFAN Study. These scans were also segmented for ground truth by two experts blind to all other data. Comparisons between machine vs expert total pericardial volume and EAT volume were made using Lin's concordance correlation coefficient (CCC). The algorithm was then applied externally in 1588 CCTAs from the SCOTHEART trial (UK), and the EAT volume was automatically calculated for each case. Cross-sectional associations between standardised EAT volumes and prevalent AF and CAD were performed. Results Within both the internal (UK ORFAN sites) and external (USA ORFAN sites) validation cohorts correlation between human and machine segmented total pericardium and EAT was excellent, with CCC of 0.97 for both volumes (external validation cohort shown in Figure 2A). Utilising SCOTHEART CCTAs with automatically segmented EAT volumes, a multivariable-adjusted logistic regression model accounting for risk factors of age, sex, BMI, hypertension, diabetes mellitus, valvular disease, and previous heart surgery found that EAT volumes were significantly associated with prevalent AF, with odds ratio (OR) per 1 SD increase of EAT volume of 1.20 (95% CI, 1.06 to 1.44; P=0.03). A similar model for prevalent CAD, adjusted for age, sex, BMI, hypertension, non-HDL cholesterol, diabetes mellitus, and coronary artery calcium score resulted in an OR per 1 SD increase of EAT volume of 1.26 (95% CI, 1.10 to 1.45; P=0.001) (Figure 2B). Conclusion Highly accurate, reproducible, and instantaneous EAT volume quantification is possible utilising deep-learning detection of the whole human heart within the pericardial sac. Funding Acknowledgement Type of funding sources: Public Institution(s). Main funding source(s): British Heart FoundationNational Institute for Health Research - Oxford University Hospitals Biomedical Research Centre Figure 1Figure 2

Convolutional neural network-based automatic heart segmentation and quantitation in 123I-metaiodobenzylguanidine SPECT imaging

BackgroundSince three-dimensional segmentation of cardiac region in 123I-metaiodobenzylguanidine (MIBG) study has not been established, this study aimed to achieve organ segmentation using a convolutional neural network (CNN) with 123I-MIBG single photon emission computed tomography (SPECT) imaging, to calculate heart counts and washout rates (WR) automatically and to compare with conventional quantitation based on planar imaging.MethodsWe assessed 48 patients (aged 68.4 ± 11.7 years) with heart and neurological diseases, including chronic heart failure, dementia with Lewy bodies, and Parkinson's disease. All patients were assessed by early and late 123I-MIBG planar and SPECT imaging. The CNN was initially trained to individually segment the lungs and liver on early and late SPECT images. The segmentation masks were aligned, and then, the CNN was trained to directly segment the heart, and all models were evaluated using fourfold cross-validation. The CNN-based average heart counts and WR were calculated and compared with those determined using planar parameters. The CNN-based SPECT and conventional planar heart counts were corrected by physical time decay, injected dose of 123I-MIBG, and body weight. We also divided WR into normal and abnormal groups from linear regression lines determined by the relationship between planar WR and CNN-based WR and then analyzed agreement between them.ResultsThe CNN segmented the cardiac region in patients with normal and reduced uptake. The CNN-based SPECT heart counts significantly correlated with conventional planar heart counts with and without background correction and a planar heart-to-mediastinum ratio (R2 = 0.862, 0.827, and 0.729, p < 0.0001, respectively). The CNN-based and planar WRs also correlated with and without background correction and WR based on heart-to-mediastinum ratios of R2 = 0.584, 0.568 and 0.507, respectively (p < 0.0001). Contingency table findings of high and low WR (cutoffs: 34% and 30% for planar and SPECT studies, respectively) showed 87.2% agreement between CNN-based and planar methods.ConclusionsThe CNN could create segmentation from SPECT images, and average heart counts and WR were reliably calculated three-dimensionally, which might be a novel approach to quantifying SPECT images of innervation.

Bidirectional cross-modality unsupervised domain adaptation using generative adversarial networks for cardiac image segmentation.

A novel Generative Adversarial Networks (GAN) based bidirectional cross-modality unsupervised domain adaptation (GBCUDA) framework is developed for cardiac image segmentation, which can effectively tackle the problem of network's segmentation performance degradation when adapting to the target domain without ground truth labels. GBCUDA uses GAN for image alignment, applies adversarial learning to extract image features, and gradually enhances the domain invariance of extracted features. The shared encoder performs an end-to-end learning task in which features that differ between the two domains complement each other. The self-attention mechanism is incorporated to the GAN network, which can generate details based on the prompts of all feature positions. Furthermore, spectrum normalization is implemented to stabilize the training of GAN, and knowledge distillation loss is introduced to process high-level feature-maps in order to better complete the cross-mode segmentation task. The effectiveness of our proposed unsupervised domain adaptation framework is tested over the Multi-Modality Whole Heart Segmentation (MM-WHS) Challenge 2017 dataset. The proposed method is able to improve the average Dice from 74.1% to 81.5% for the four cardiac substructures, and reduce the average symmetric surface distance (ASD) from 7.0 to 5.8 over CT images. For MRI images, our proposed framework trained on CT images gives the average Dice of 59.2% and reduces the average ASD from 5.7 to 4.9. The evaluation results demonstrate our method's effectiveness on domain adaptation and the superiority to the current state-of-the-art domain adaptation methods.

Validation of a Whole Heart Segmentation from Computed Tomography Imaging Using a Deep-Learning Approach.

The aim of this study is to develop an automated deep-learning-based whole heart segmentation of ECG-gated computed tomography data. After 21 exclusions, CT acquired before transcatheter aortic valve implantation in 71 patients were reviewed and randomly split in a training (n = 55 patients), validation (n = 8 patients), and a test set (n = 8 patients). A fully automatic deep-learning method combining two convolutional neural networks performed segmentation of 10 cardiovascular structures, which was compared with the manually segmented reference by the Dice index. Correlations and agreement between myocardial volumes and mass were assessed. The algorithm demonstrated high accuracy (Dice score = 0.920; interquartile range: 0.906-0.925) and a low computing time (13.4s, range 11.9-14.9). Correlations and agreement of volumes and mass were satisfactory for most structures. Six of ten structures were well segmented. Deep-learning-based method allowed automated WHS from ECG-gated CT data with a high accuracy. Challenges remain to improve right-sided structures segmentation and achieve daily clinical application.

Evaluation of the feasibility of explainable computer-aided detection of cardiomegaly on chest radiographs using deep learning

We examined the feasibility of explainable computer-aided detection of cardiomegaly in routine clinical practice using segmentation-based methods. Overall, 793 retrospectively acquired posterior–anterior (PA) chest X-ray images (CXRs) of 793 patients were used to train deep learning (DL) models for lung and heart segmentation. The training dataset included PA CXRs from two public datasets and in-house PA CXRs. Two fully automated segmentation-based methods using state-of-the-art DL models for lung and heart segmentation were developed. The diagnostic performance was assessed and the reliability of the automatic cardiothoracic ratio (CTR) calculation was determined using the mean absolute error and paired t-test. The effects of thoracic pathological conditions on performance were assessed using subgroup analysis. One thousand PA CXRs of 1000 patients (480 men, 520 women; mean age 63 ± 23 years) were included. The CTR values derived from the DL models and diagnostic performance exhibited excellent agreement with reference standards for the whole test dataset. Performance of segmentation-based methods differed based on thoracic conditions. When tested using CXRs with lesions obscuring heart borders, the performance was lower than that for other thoracic pathological findings. Thus, segmentation-based methods using DL could detect cardiomegaly; however, the feasibility of computer-aided detection of cardiomegaly without human intervention was limited.

Spectral augmentation for heart chambers segmentation on conventional contrasted and unenhanced CT scans: an in-depth study.

Recently, machine learning has outperformed established tools for automated segmentation in medical imaging. However, segmentation of cardiac chambers still proves challenging due to the variety of contrast agent injection protocols used in clinical practice, inducing disparities of contrast between cavities. Hence, training a generalist network requires large training datasets representative of these protocols. Furthermore, segmentation on unenhanced CT scans is further hindered by the challenge of obtaining ground truths from these images. Newly available spectral CT scanners allow innovative image reconstructions such as virtual non-contrast (VNC) imaging, mimicking non-contrasted conventional CT studies from a contrasted scan. Recent publications have demonstrated that networks can be trained using VNC to segment contrasted and unenhanced conventional CT scans to reduce annotated data requirements and the need for annotations on unenhanced scans. We propose an extensive evaluation of this statement. We undertake multiple trainings of a 3D multi-label heart segmentation network with (HU-VNC) and without (HUonly) VNC as augmentation, using decreasing training dataset sizes (114, 76, 57, 38, 29, 19 patients). At each step, both networks are tested on a multi-vendor, multi-centric dataset of 122 patients, including different protocols: pulmonary embolism (PE), chest-abdomen-pelvis (CAP), heart CT angiography (CTA) and true non-contrast scans (TNC). An in-depth comparison of resulting Dice coefficients and distance metrics is performed for the networks trained on the largest dataset. HU-VNC-trained on 57 patients significantly outperforms HUonly trained on 114 regarding CAP and TNC scans (mean Dice coefficients of 0.881/0.835 and 0.882/0.416, respectively). When trained on the largest dataset, significant improvements in all labels are noted for TNC and CAP scans (mean Dice coefficient of 0.882/0.416 and 0.891/0.835, respectively). Adding VNC images as training augmentation allows the network to perform on unenhanced scans and improves segmentations on other imaging protocols, while using a reduced training dataset.

INFERIOR VENA CAVA FILTER RETRIEVAL 23 MONTH AFTER IMPLANTATION WITH STANDARD SNARE LOOP TECHNIQUE : CASE REPORT

1. Introduction. Pulmonary embolism (PE ) and Deep vein thrombosis (DVT ) continues to be a cause of signicant morbidity and mortality which is also third 1 leading cardiovascular cause of death . If not treated properly it also leads to chronic pulmonary hypertension and cor pulmonale . Once DVT is diagnosed treatmet is required and consist of oral anticoagulation and inferior vena cava interruption via lter placement in patients with contraindication for anticoagulation treatment. In 2003 ,the U.S .Food and Drug Administration (FDA ) cleared retrievable inferior vena cave lters (rIVCFs) for clinical use , after that its uses exponentially increases . In parallel with these trends ,there has been growing awareness of device –related complication including fracture , penetration to adjuscent structure , migration , thrombosis and some time embolisation of fractured segment in heart that leads to cardiac 2 3 peforation , arrhythmia and death . In response , the FDA released safety communication in 2010 and 2014 advocating immediate retrieval of 4 lter once mechanical prophylaxis was no longer uses .However retrieval of lter remained rare with rates reported as low as 8.5 %. . Once lter remain in place for long duration, normal standard snare retrieval technique usually not work and alternative techniques like , Sling technique , 5 Excimer laser and Endobronchial forcep supported methods can be used . But prolong duration of lter implantion always associated with less chance of lter retrieval and associated with multiple complication .

Convolutional neural network for semantic segmentation of fetal echocardiography based on four-chamber view

The acute shortage of trained and experienced sonographers causes the detection of congenital heart defects (CHDs) extremely difficult. In order to minimize this difficulty, an accurate fetal heart segmentation to the early location of such structural heart abnormalities prior to delivery is essential. However, the segmentation process is not an easy task due to the small size of the fetal heart structure. Moreover, the manual task for identifying the standard cardiac planes, primarily based on a four-chamber view, requires a well-trained clinician and experience. In this paper, a CNN method using U-Net architecture was proposed to automate fetal cardiac standard planes segmentation from ultrasound images. A total of 519 fetal cardiac images was obtained from three videos. All data is divided into training and testing data. The testing data consist of 106 slices of the four-chamber segmentation tasks, i.e. atrial septal defect (ASD), ventricular septal defect (VSD), and normal. The segmentation of the post-processing method is needed to enhanced the segmentation result. In this paper, a combination technique with U-Net and Otsu thresholding gives the best performances with 99.48%-pixel accuracy, 96.73% mean accuracy, 94.92% mean intersection over union, and 0.21% segmentation error. In the future, the implementation of Deep Learning in the study of CHDs holds significant potential for identifying novel CHDs in heterogeneous fetal hearts.

Causes of altered ventricular mechanics in hypertrophic cardiomyopathy: an in-silico study

BackgroundHypertrophic cardiomyopathy (HCM) is typically caused by mutations in sarcomeric genes leading to cardiomyocyte disarray, replacement fibrosis, impaired contractility, and elevated filling pressures. These varying tissue properties are associated with certain strain patterns that may allow to establish a diagnosis by means of non-invasive imaging without the necessity of harmful myocardial biopsies or contrast agent application. With a numerical study, we aim to answer: how the variability in each of these mechanisms contributes to altered mechanics of the left ventricle (LV) and if the deformation obtained in in-silico experiments is comparable to values reported from clinical measurements.MethodsWe conducted an in-silico sensitivity study on physiological and pathological mechanisms potentially underlying the clinical HCM phenotype. The deformation of the four-chamber heart models was simulated using a finite-element mechanical solver with a sliding boundary condition to mimic the tissue surrounding the heart. Furthermore, a closed-loop circulatory model delivered the pressure values acting on the endocardium. Deformation measures and mechanical behavior of the heart models were evaluated globally and regionally.ResultsHypertrophy of the LV affected the course of strain, strain rate, and wall thickening—the root-mean-squared difference of the wall thickening between control (mean thickness 10 mm) and hypertrophic geometries (17 mm) was >10%. A reduction of active force development by 40% led to less overall deformation: maximal radial strain reduced from 26 to 21%. A fivefold increase in tissue stiffness caused a more homogeneous distribution of the strain values among 17 heart segments. Fiber disarray led to minor changes in the circumferential and radial strain. A combination of pathological mechanisms led to reduced and slower deformation of the LV and halved the longitudinal shortening of the LA.ConclusionsThis study uses a computer model to determine the changes in LV deformation caused by pathological mechanisms that are presumed to underlay HCM. This knowledge can complement imaging-derived information to obtain a more accurate diagnosis of HCM.

Dark blood ischemic LGE segmentation using a deep learning approach

Abstract Funding Acknowledgements Type of funding sources: None. Introduction The extent of ischemic scar detected by Cardiac Magnetic Resonance (CMR) with late gadolinium enhancement (LGE) is linked with long-term prognosis, but scar quantification is time-consuming. Deep Learning (DL) approaches appear promising in CMR segmentation. Purpose: To train and apply a deep learning approach to dark blood (DB) CMR-LGE for ischemic scar segmentation, comparing results to 4-Standard Deviation (4-SD) semi-automated method. Methods: We trained and validated a dual neural network infrastructure on a dataset of DB-LGE short-axis stacks, acquired at 1.5T from 33 patients with ischemic scar. The DL architectures were an evolution of the U-Net Convolutional Neural Network (CNN), using data augmentation to increase generalization. The CNNs worked together to identify and segment 1) the myocardium and 2) areas of LGE. The first CNN simultaneously cropped the region of interest (RoI) according to the bounding box of the heart and calculated the area of myocardium. The cropped RoI was then processed by the second CNN, which identified the overall LGE area. The extent of scar was calculated as the ratio of the two areas. For comparison, endo- and epi-cardial borders were manually contoured and scars segmented by a 4-SD technique with a validated software. Results: The two U-Net networks were implemented with two free and open-source software library for machine learning. We performed 5-fold cross-validation over a dataset of 108 and 385 labelled CMR images of the myocardium and scar, respectively. We obtained high performance (&amp;gt; ∼0.85) as measured by the Intersection over Union metric (IoU) on the training sets, in the case of scar segmentation. With regards to heart recognition, the performance was lower (&amp;gt; ∼0.7), although improved (∼ 0.75) by detecting the cardiac area instead of heart boundaries. On the validation set, performances oscillated between 0.8 and 0.85 for scar tissue recognition, and dropped to ∼0.7 for myocardium segmentation. We believe that underrepresented samples and noise might be affecting the overall performances, so that additional data might be beneficial. Figure1: examples of heart segmentation (upper left panel: training; upper right panel: validation) and of scar segmentation (lower left panel: training; lower right panel: validation). Conclusion: Our CNNs show promising results in automatically segmenting LV and quantify ischemic scars on DB-LGE-CMR images. The performances of our method can further improve by expanding the data set used for the training. If implemented in a clinical routine, this process can speed up the CMR analysis process and aid in the clinical decision-making. Abstract Figure.

Application of an automatic segmentation method for evaluating cardiac structure doses received by breast radiotherapy patients.

Quantifying radiation dose to cardiac substructures is important for research on the etiology and prevention of complications following radiotherapy; however, segmentation of substructures is challenging. In this study we demonstrate the application of our atlas-based automatic segmentation method to breast cancer radiotherapy plans for generating radiation doses in support of late effects research. We applied our segmentation method to contour heart substructures on the computed tomography (CT) images of 70 breast cancer patients who received external photon radiotherapy. Two cardiologists provided manual segmentation of the whole heart (WH), left/right atria, left/right ventricles, and left anterior descending artery (LAD). The automatically contours were compared with manual delineations to evaluate similarity in terms of geometry and dose. The mean Dice similarity coefficient between manual and automatic segmentations was 0.96 for the WH, 0.65 to 0.82 for the atria and ventricles, and 0.06 for the LAD. The mean average surface distance was 1.2mm for the WH, 3.4 to 4.1mm for the atria and ventricles, and 6.4mm for the LAD. We found the dose to the cardiac substructures based on our automatic segmentation agrees with manual segmentation within expected observer variability. For left breast patients, the mean absolute difference in mean dose was 0.1Gy for the WH, 0.2 to 0.7Gy for the atria and ventricles, and 1.8Gy for the LAD. For right breast patients, these values were 0.0Gy, 0.1 to 0.4Gy, and 0.4Gy, respectively. Our automatic segmentation method will facilitate the development of radiotherapy prescriptive criteria for mitigating cardiovascular complications.

Prediction of Coronary Calcification and Stenosis: Role of Radiomics From Low-Dose CT

We aimed to assess relationship between single-click, whole heart radiomics from low-dose computed tomography (LDCT) for lung cancer screening with coronary artery calcification and stenosis. The institutional review board-approved, retrospective study included all 106 patients (68 men, 38 women, mean age 64 ± 7 years) who underwent both LDCT for lung cancer screening and had calcium scoring and coronary computed tomography angiography in our institution. We recorded the clinical variables including patients' demographics, smoking history, family history, and lipid profiles. Coronary calcium scores and grading of coronary stenosis were recorded from the radiology information system. We calculated the multiethnic scores for atherosclerosis risk scores to obtain 10-year coronary heart disease (MESA 10-Y CHD) risk of cardiovascular disease for all patients. Deidentified LDCT exams were exported to a Radiomics prototype for automatic heart segmentation, and derivation of radiomics. Data were analyzed using multiple logistic regression and kernel Fisher discriminant analyses. Whole heart radiomics were better than the clinical variables for differentiating subjects with different Agatston scores (≤400 and >400) (area under the curve [AUC] 0.92 vs 0.69). Prediction of coronary stenosis and MESA 10-Y CHD risk was better on whole heart radiomics (AUC:0.86-0.87) than with clinical variables (AUC:0.69-0.79). Addition of clinical variables or visual assessment of coronary calcification from LDCT to whole heart radiomics resulted in a modest change in the AUC. Single-click, whole heart radiomics obtained from LDCT for lung cancer screening can differentiate patients with different Agatston and MESA risk scores for cardiovascular diseases.

Model-based segmentation using neural network-based boundary detectors: Application to prostate and heart segmentation in MR images

Model-based segmentation (MBS) is a variant of active surfaces and active shape models that has successfully been used to segment anatomical structures such as the heart or the brain. We propose to integrate neural networks (NNs) into MBS for boundary detection. We formulate boundary detection as a regression task and use a NN to predict the distances between a surface mesh and the corresponding boundary points. The proposed approach has been applied to two tasks — prostate segmentation in MR images and the segmentation of the left and right ventricle in MR images. For the first task, data from the Prostate MR Image Segmentation 2012 (PROMISE12) challenge has been used. For the second task, a diverse database with cardiac MR images from six clinical sites has been used. We compare the results to the popular U-net approaches using the nnU-net implementation that is among the top performing segmentation algorithms in various challenges. In cross-validation experiments, the mean Dice scores are very similar and no statistically significant difference is observed. On the PROMISE12 test set, nnU-net Dice scores are significantly better. This is achieved by using an ensemble of 2D and 3D U-nets to generate the final segmentation, a concept that may also be adapted to NN-based boundary detection in the future. While the U-net provides a voxel labeling, our approach provides a 3D surface mesh with pre-defined mesh topology, establishes correspondences with respect to the reference mesh, avoids isolated falsely segmented regions and ensures proper connectivity of different regions.

Convolutional neural network for semantic segmentation of fetal echocardiography based on four-chamber view

The acute shortage of trained and experienced sonographers causes the detection of congenital heart defects (CHDs) extremely difficult. In order to minimize this difficulty, an accurate fetal heart segmentation to the early location of such structural heart abnormalities prior to delivery is essential. However, the segmentation process is not an easy task due to the small size of the fetal heart structure. Moreover, the manual task for identifying the standard cardiac planes, primarily based on a four-chamber view, requires a well-trained clinician and experience. In this paper, a CNN method using U-Net architecture was proposed to automate fetal cardiac standard planes segmentation from ultrasound images. A total of 519 fetal cardiac images was obtained from three videos. All data is divided into training and testing data. The testing data consist of 106 slices of the four-chamber segmentation tasks, i.e. atrial septal defect (ASD), ventricular septal defect (VSD), and Normal. The segmentation of the post-processing method is needed to enhanced the segmentation result. In this paper, a combination technique with U-Net and Otsu thresholding gives the best performances with 99.48%-pixel accuracy, 96.73% mean accuracy, 94.92% mean intersection over union, and 0.21% segmentation error. In the future, the implementation of Deep Learning in the study of CHDs holds significant potential for identifying novel CHDs in heterogeneous fetal hearts.

Development of the single ventricle heart mathematical model based on the equation of forced oscillations

Not only protozoa being the simplest organisms have a single ventricle heart. Artificial heart systems used in cardiac surgery can be also of the single-ventricle design. Besides, there is a “single-ventricle heart condition”, i.e. a complex of defects in the structure of the heart that may occur in some human individuals and that may be found with different occurrence rate and severity. This defect should be attributed to an uneven development and malformation of some individual heart segments in the prenatal period. Single Ventricle belongs to the category of congenital heart defects. The latter greatly increases the scientific interest in studying the features of mathematical models of the single-ventricle heart condition. Herein, we offer our study on some issues of the smoothness of sewing of the exact solutions of a mathematical model of a single-ventricle heart based on the equation of forced oscillations.

Whole Heart Segmentation Using 3D FM-Pre-ResNet Encoder–Decoder Based Architecture with Variational Autoencoder Regularization

An accurate whole heart segmentation (WHS) on medical images, including computed tomography (CT) and magnetic resonance (MR) images, plays a crucial role in many clinical applications, such as cardiovascular disease diagnosis, pre-surgical planning, and intraoperative treatment. Manual whole-heart segmentation is a time-consuming process, prone to subjectivity and error. Therefore, there is a need to develop a quick, automatic, and accurate whole heart segmentation systems. Nowadays, convolutional neural networks (CNNs) emerged as a robust approach for medical image segmentation. In this paper, we first introduce a novel connectivity structure of residual unit that we refer to as a feature merge residual unit (FM-Pre-ResNet). The proposed connectivity allows the creation of distinctly deep models without an increase in the number of parameters compared to the pre-activation residual units. Second, we propose a three-dimensional (3D) encoder–decoder based architecture that successfully incorporates FM-Pre-ResNet units and variational autoencoder (VAE). In an encoding stage, FM-Pre-ResNet units are used for learning a low-dimensional representation of the input. After that, the variational autoencoder (VAE) reconstructs the input image from the low-dimensional latent space to provide a strong regularization of all model weights, simultaneously preventing overfitting on the training data. Finally, the decoding stage creates the final whole heart segmentation. We evaluate our method on the 40 test subjects of the MICCAI Multi-Modality Whole Heart Segmentation (MM-WHS) Challenge. The average dice values of whole heart segmentation are 90.39% (CT images) and 89.50% (MRI images), which are both highly comparable to the state-of-the-art.