Deep Learning for Segmentation Using an Open Large-Scale Dataset in 2D Echocardiography.

Interests have been rapidly growing in the field of radiotherapy to replace CT with magnetic resonance imaging (MRI), due to superior soft tissue contrast offered by MRI and the desire to reduce unnecessary radiation dose. MR-only radiotherapy also simplifies clinical workflow and avoids uncertainties in aligning MR with CT. Methods, however, are needed to derive CT-equivalent representations, often known as synthetic CT (sCT), from patient MR images for dose calculation and DRR-based patient positioning. Synthetic CT estimation is also important for PET attenuation correction in hybrid PET-MR systems. We propose in this work a novel deep convolutional neural network (DCNN) method for sCT generation and evaluate its performance on a set of brain tumor patient images. The proposed method builds upon recent developments of deep learning and convolutional neural networks in the computer vision literature. The proposed DCNN model has 27 convolutional layers interleaved with pooling and unpooling layers and 35 million free parameters, which can be trained to learn a direct end-to-end mapping from MR images to their corresponding CTs. Training such a large model on our limited data is made possible through the principle of transfer learning and by initializing model weights from a pretrained model. Eighteen brain tumor patients with both CT and T1-weighted MR images are used as experimental data and a sixfold cross-validation study is performed. Each sCT generated is compared against the real CT image of the same patient on a voxel-by-voxel basis. Comparison is also made with respect to an atlas-based approach that involves deformable atlas registration and patch-based atlas fusion. The proposed DCNN method produced a mean absolute error (MAE) below 85 HU for 13 of the 18 test subjects. The overall average MAE was 84.8 ± 17.3 HU for all subjects, which was found to be significantly better than the average MAE of 94.5 ± 17.8 HU for the atlas-based method. The DCNN method also provided significantly better accuracy when being evaluated using two other metrics: the mean squared error (188.6 ± 33.7 versus 198.3 ± 33.0) and the Pearson correlation coefficient(0.906 ± 0.03 versus 0.896 ± 0.03). Although training a DCNN model can be slow, training only need be done once. Applying a trained model to generate a complete sCT volume for each new patient MR image only took 9 s, which was much faster than the atlas-based approach. A DCNN model method was developed, and shown to be able to produce highly accurate sCT estimations from conventional, single-sequence MR images in near real time. Quantitative results also showed that the proposed method competed favorably with an atlas-based method, in terms of both accuracy and computation speed at test time. Further validation on dose computation accuracy and on a larger patient cohort is warranted. Extensions of the method are also possible to further improve accuracy or to handle multi-sequence MR images.

Recommendations for cardiac chamber quantification by echocardiography in adults: an update from the American Society of Echocardiography and the European Association of Cardiovascular Imaging.

Point cloud filtering is an important prerequisite for three-dimensional surface modeling with high precision based on LiDAR data. To cope with the issues of low filtering accuracy or excessive model complexity in traditional filtering algorithms, this paper proposes a filtering method for LiDAR point cloud based on a multi-scale convolutional neural network incorporated with the attention mechanism. Firstly, a regular image patch centering on each point is constructed based on the elevation information of point clouds. As thus, the point cloud filtering problem is transformed into the image classification problem. Then, considering the ability of multi-scale convolution to extract features at different scales and the potential of the attention mechanism to capture key information in images, a multi-scale convolutional neural network framework is constructed, and the attention mechanism is incorporated to coordinate multi-scale convolution kernel with channel and spatial attention modules. After this, the feature maps of the LiDAR point clouds can be acquired at different scales. For these feature maps, the weights of each channel layer and different spatial regions can be further tuned adaptively, which makes the network training more targeted, thereby improving the model performance for image classification and eventually separating of ground points and non-ground points preferably. Finally, the proposed method is compared with the cloth simulation filtering method (CSF), deep neural network method (DNN), k-nearest neighbor method (KNN), deep convolutional neural network method (DCNN) and scale-irrelevant and terrain-adaptive method (SITA) for the standard ISPRS dataset of point cloud filtering and the filter dataset of Qinghai. The experimental results show that the proposed method can obtain lower classification errors, which proves the superiority of this method in point cloud filtering.

Underwater images often suffer from scattering and color distortion because of underwater light transportation characteristics and water impurities. Presence of such factors make underwater image classification task very challenging. We propose a novel classification convolution autoencoder (CCAE), which can classify large size underwater images with promising accuracy. CCAE is designed as a hybrid network, which combines benefits of unsupervised convolution autoencoder to extract non-trivial features and a classifier, for better classification accuracy. In order to evaluate classification accuracy of proposed network, experiments are conducted on Fish4Knowledge dataset and underwater synsets of benchmark ImageNet dataset. Classification accuracy, precision, recall and f1-score results are compared with state-of-the-art deep convolutional neural network (CNN) methods. Results show that proposed system can accurately classify large-size underwater images with promising accuracy and outperforms state-of-the-art deep CNN methods. With the proposed network, we expect to advance underwater image classification research and its applications in many areas like ocean biology, sea exploration and aquatic robotics.

Quantitative analysis of left-ventricular (LV) aneurysms after myocardial infarction is prognostically relevant and assists in planning surgery. Three-dimensional (3D) echocardiography facilitates clear visualization of cardiac anatomy and accurate assessment of functional parameters. The aim of the present study was to determine the ability of 3D echocardiography to quantify LV aneurysms. Ten patients with a known LV-aneurysm after myocardial infarction underwent 3D echocardiography and cardiac magnetic resonance (CMR) imaging at 1.5 Tesla within 3 days. For 3D echocardiography, a multiplanar transesophageal examination was performed with full LV coverage and the 3D dataset was analyzed offline. The LV-aneurysm was defined by a wall thickness <5 mm. The following quantitative parameters were determined: left ventricular end-diastolic and end-systolic volumes, LV myocardial mass (LV-mass) and mass of the LV-aneurysm. LV ejection fraction and percentage of aneurysm mass (%-aneurysm) were calculated. LV volumes and ejection fraction showed a strong correlation between 3D echocardiography and CMR (r = 0.94-0.97; P < 0.01). Importantly, the mass and percentage of mass of the LV-aneurysm demonstrated a high correlation as well (r = 0.94 and r = 0.86, respectively; P < 0.01). For all parameters, the calculated bias between both methods was found to be minimal (0.8-7.6%). Three-dimensional echocardiography proved to be a reliable tool for quantitative analysis of LV volumes, ejection fraction and aneurysm size in patients with prior myocardial infarction. In addition, 3D visualization of the complex cardiac anatomy in patients with LV-aneurysm may assist surgical procedure planning.

Utility of Three-Dimensional and Four-Dimensional Transesophageal Echocardiography in Decision-Making in a Patient with Iatrogenic Left Ventricle–to–Right Atrium Shunt (Gerbode Defect)

EAE/ASE Recommendations for Image Acquisition and Display Using Three-Dimensional Echocardiography

Selection of patients who are viable candidates for cardiac resynchronization therapy (CRT), prediction of the response to CRT as well as an optimal definition of a favorable response, all require further exploration. The purpose of this study was to evaluate the interplay between the prediction of the response to CRT and the definition of a favorable outcome. Seventy patients who received CRT were included. All patients met current guideline criteria for CRT. Forty-three echocardiographic parameters were evaluated before CRT and at 1, 3, 6, and 12months. M-mode, 2D echocardiography, and Doppler imaging were used to quantify left ventricular (LV) systolic and diastolic function, mitral regurgitation, right ventricular systolic function, pulmonary artery pressure, and myocardial mechanical dyssynchrony. The following definitions of a favorable CRT response were used: left ventricular ejection fraction (LVEF) improvement more >5% acutely following CRT, LVEF improvement >20% at 12-month follow-up, and a LV end-systolic volume (LVESV) decrease >15% at 12-month follow-up. For the LVEF improvement >5%, the best predictor was isovolumetric relaxation time (IVRT; P=.035). For improvement of LVEF >20%, the best predictors were left ventricular stroke index (LVSI; P=.044) and left ventricular fractional shortening (LVFS; P=.031). For the drop in left ventricular systolic volume (LVESV >15%), the best predictor was septal-to-lateral wall delay (ΔT) (P=.043, RR=1.023, 95% CI for RR=1.001-1.045). The definition of a favorable CRT response influenced the optimal predictor variable(s). Standardization of defining a favorable response to CRT is needed to guide clinical decision making processes.

To compare two-dimensional (2D) echocardiography, the current method of screening for treatment-related cardiomyopathy recommended by the Children's Oncology Group Guidelines, to cardiac magnetic resonance (CMR) imaging, the reference standard for left ventricular (LV) function. Cross-sectional, contemporaneous evaluation of LV structure and function by 2D and three-dimensional (3D) echocardiography and CMR imaging in 114 adult survivors of childhood cancer currently median age 39 years (range, 22 to 53 years) exposed to anthracycline chemotherapy and/or chest-directed radiation therapy. In this survivor population, 14% (n = 16) had an ejection fraction (EF) less than 50% by CMR. Survivors previously undiagnosed with cardiotoxicity (n = 108) had a high prevalence of EF (32%) and cardiac mass (48%) that were more than two standard deviations below the mean of normative CMR data. 2D echocardiography overestimated the mean EF of this population by 5%. Compared with CMR, 2D echocardiography (biplane method) had a sensitivity of 25% and a false-negative rate of 75% for detection of EF less than 50%, although 3D echocardiography had 53% and 47%, respectively. Twelve survivors (11%) had an EF less than 50% by CMR but were misclassified as ≥ 50% (range, 50% to 68%) by 2D echocardiography (biplane method). Detection of cardiomyopathy was improved (sensitivity, 75%) by using a higher 2D echocardiography cutoff (EF < 60%) to detect an EF less than 50% by the reference standard CMR. CMR identified a high prevalence of cardiomyopathy among adult survivors previously undiagnosed with cardiac disease. 2D echocardiography demonstrated limited screening performance. In this high-risk population, survivors with an EF 50% to 59% by 2D echocardiography should be considered for comprehensive cardiac assessment, which may include CMR.

3D Echocardiography: A Review of the Current Status and Future Directions

Background. Multi-slice computed tomography (MSCT) has been demonstrated as a feasible imaging modality for non-invasive assessment of coronary artery disease and left ventricular (LV) function analysis. Recently, 320-slice systems have become available with 16 cm anatomical coverage allowing prospective image acquisition of the entire heart within a single rotation or heart beat. However, limited data are currently available with these systems. The purpose of the present study therefore was to evaluate the accuracy of 320-slice MSCT in the assessment of global LV function as compared to 2-dimensional (2D) echocardiography. Methods. A head-to-head comparison between 320-slice MSCT and 2D echocardiography was performed in 40 patients (24 male; mean age 61 ± 9 years) with known or suspected coronary artery disease (CAD). During intravenous contrast agent administration, the entire heart was imaged in a single heartbeat, using prospective dose modulation (full dose during 65–85% of R-R interval). The following parameters were used: gantry rotation time 350 ms, tube voltage 120 kV, tube current 300–500mA. LV end-diastolic volumes (LVEDV) and LV end-systolic volumes (LVESV) were determined and the LV ejection fraction (LVEF) was derived. Two-dimensional echocardiography served as the gold standard. Results. Average LVEF was 59% ± 8% (range 31%–77%) as determined on 2D-echocardiography, compared with 61% ± 8% (range 33%–78%) on MSCT. Evaluation of LVEF by linear regression analysis showed a good correlation between MSCT and 2D-echocardiography (r = 0.84; p &lt; 0.001). A close correlation between MSCT and 2D-echocardiography was also demonstrated for the assessment of LVEDV (r = 0.81; p &lt; 0.001) and LVESV (r = 0.89; p &lt; 0.001). At Bland-Altman analysis, mean differences (± SD) of 20.88 ml ± 23.07 ml (p &lt; 0.01) and 6.83 ml ± 13.01 ml (p &lt; 0.01) were observed between MSCT and 2D-echocardiography for LVEDV and LVESV respectively. As a result, LVEF was slightly overestimated with MSCT (1.98% ± 4.54%; p &lt; 0.01). Conclusion. Accurate assessment of LV function and volumes is feasible with 320-slice MSCT in patients with known or suspected CAD.

Assessment of global left ventricular function and volumes with 320-row multidetector computed tomography: A comparison with 2D-echocardiography

BACKGROUNDThe importance of the right ventricle (RV) has been increasingly recognized, and accurate RV measurement has become necessary. However, assessment of the RV with two-dimensional (2D) echocardiography has several limitations. As the development of novel methods for RV measurement continues, we can expect more accordant values related to RV geometry.METHODSFifty-eight subjects who were examined by transthoracic echocardiography (TTE) immediately before and after hemodialysis (HD) were enrolled. Real-time, full-volume, three-dimensional (3D) echocardiographic images were acquired and analyzed using dedicated software. Conventional RV parameters for RV size and function were measured for comparison with pre-HD and post-HD values by both 2D-TTE and 3D-TTE.RESULTS3D RV volumes and ejection fractions were significantly decreased after HD. The values of the 3D image-derived RV dimensions, tricuspid annular plane systolic excursion (TAPSE), fractional area change (FAC), and longitudinal strain were also affected by acute preload changes (TAPSE pre-HD: 22.4 ± 4.0 mm, post-HD: 19.0 ± 4.2 mm, p < 0.001; FAC pre-HD: 49.6% ± 5.9%, post-HD: 46.4% ± 5.5%, p < 0.001; septal longitudinal strain pre-HD: -20.1% ± 3.7%, post-HD: -16.8% ± 3.8%, p < 0.001). With the exception of FAC, most 2D RV parameters were well correlated with the 3D values.CONCLUSIONSVarious parameters representing RV anatomy and function were acquired easily and more accurately from 3D echocardiographic images than from 2D images but were affected by acute preload changes. 3D TTE could be a new modality for assessing RV function and size, but each value from 3D TTE should be interpreted with caution while considering the loading condition of the patients.

Clinical applicability of conventional ultrasonographic systems using mechanical adapters for 3D echocardiographic imaging has been limited by long acquisition and processing times. We developed a rapid (6-s) acquisition technique that collects apical tomograms using a continuously internally rotating transthoracic transducer. This study was performed to examine the clinical feasibility of rapid-acquisition 3D echocardiography to estimate left ventricular end-diastolic and end-systolic volumes using electron-beam computed tomography as the reference standard. Methods and Results-We collected a series of 6 to 11 apical echocardiographic tomograms, depending on heart rate, in 11 patients. There was good correlation, low variability, and low bias between rapid 3D echocardiography and electron-beam computed tomography for measuring left ventricular end-diastolic volume (r=0.96; standard error of the estimate, 21.34 mL; bias, -4.93 mL) and left ventricular end-systolic volume (r=0.96; standard error of the estimate, 14.78 mL; bias, -6.97 mL). The rapid-acquisition 3D echocardiography extends the use of a multiplane, internally rotating handheld transducer so that it becomes a precise and clinically feasible tool for assessing left ventricular volumes and function. A rapid-image acquisition time of 6 s would allow repeated image collection during the course of a clinical echocardiographic examination. Additional work must address rapid and automated data processing.

Real-time three-dimensional (3D) echocardiography allows us to measure right ventricular (RV) end-diastolic volume irrespective of its shape. Tissue Doppler imaging (TDI) and speckle tracking imaging (STI) are new tools to assess myocardial function. We sought to evaluate RV function by 3D echocardiography and myocardial strain imaging in adult patients with atrial septal defect (ASD) before and 6 months after transcatheter closure in order to assess the utility of these new indexes in comparison with standard two-dimensional (2D) and Doppler parameters. Thirty-nine ASD patients and 39 healthy age- and sex-matched controls were studied using a commercially available cardiovascular ultrasound system. 2D-Doppler parameters of RV function (fractional area change, tricuspid annular plane systolic excursion, myocardial performance index) were calculated. 3D RV volumes were also obtained. RV peak-systolic velocities, peak-systolic strain, and peak systolic and diastolic strain-rate were measured in the basal, mid and apical segments of lateral and septal walls in apical 4-chamber view by TDI and STI. In open ASD, RV ejection fraction (3D-RVEF) and global and regional RV longitudinal strain were significantly higher than control group and decreased significantly after closure. By multivariate analysis 3D-RVEF, apical strain and strain rate were independent predictors of functional class. ROC analysis showed 3D-RVEF and apical strain to be more sensitive predictors of unfavorable outcome after defect closure compared to 2D-Doppler indexes. 3D echocardiography and myocardial strain imaging give useful insights in the quantitative assessment of RV function in ASD patients before and after closure.