In this paper, we compare common Unet architectures with upgrade version Unet++ for segmentation of the the left ventricular (LV) endocardial boundaries on ultrasound images, which were obtained by ultrasound studies of the heart. The key difference of Unet++ from Unet is the improved skip connections between encoder and decoder by adding convolution blocks. Such implementation allows us to transfer information from encoder to decoder with avoiding to loss of small details of the ultrasound image. It is very important to detect internal boundaries accuratly. We evaluated Unet++ in comparison Unet and wide Unet. The Dice coefficient for the LV segmentation with Unet++ is 91.20%. This result of Unet++ for the LV area segmentation outperforms Unet and wide Unet architectures.

Nowadays ultrasound studies of the heart, also called echocardiography (EchoCG), are widespread in modern cardiology. One of the most important steps in estimating the health of the heart is the tracking and segmentation of the left ventricular (LV) endocardial border from EchoCG, which is used for measuring the ejection fraction and assessing the regional wall motion [1]. The disadvantage of these methods is the necessity to apply image processing manually or in a semi-automatic mode, which requires special knowledge and skills. As a result, the issue of an automatic tracking and segmentation of the LV on EchoCG-images is an actual and practical problem. The capabilities of the fully trained model of the convolutional neural network Unet for automatic identification of the LV region are explored in this paper. The obtained accuracy of LV segmentation is up to 92.3%, which suggests the expediency of using Unet for automatic identification of the LV endocardial border on ultrasound images.

The paper discusses the results of the research and development of an innovative deep learning-based computer vision system for the fully automatic asbestos content (productivity) estimation in rock chunk (stone) veins in an open pit and within the time comparable with the work of specialists (about 10 min per one open pit processing place). The discussed system is based on the applying of instance and semantic segmentation of artificial neural networks. The Mask R-CNN-based network architecture is applied to the asbestos-containing rock chunks searching images of an open pit. The U-Net-based network architecture is applied to the segmentation of asbestos veins in the images of selected rock chunks. The designed system allows an automatic search and takes images of the asbestos rocks in an open pit in the near-infrared range (NIR) and processes the obtained images. The result of the system work is the average asbestos content (productivity) estimation for each controlled open pit. It is validated to estimate asbestos content as the graduated average ratio of the vein area value to the selected rock chunk area value, both determined by the trained neural network. For both neural network training tasks the training, validation, and test datasets are collected. The designed system demonstrates an error of about 0.4% under different weather conditions in an open pit when the asbestos content is about 1.5–4%. The obtained accuracy is sufficient to use the system as a geological service tool instead of currently applied visual-based estimations.

One of the most significant tasks of echocardiography is the automatic delineation of the cardiac structures from 2D echocardiographic images. Over the past decades, the automation of this task has been the subject of intense research. One of the most effective approaches is based on the deep convolutional neural networks. Nonetheless, it is necessary to use echocardiogram frames of the cardiac muscle, which show the boundaries of the cardiac structures labeled/annotated by experts/cardiologists to train it. However, the number of databases containing the necessary information is relatively small. Therefore, generated echocardiogram frames are used to increase the amount of training samples. This process is based on the ultrasound images of the heart, annotated by experts. The article proposes an improved method for generating echocardiograms using a generative adversarial neural network (GAN) with a patch-based conditional discriminator. It has been demonstrated that it is possible to improve the quality of generated echocardiogram frames in both two and four chamber views (AP4C, AP2C) using the masks of cardiac segmentation with sub-pixel convolution layer (pixel shuffle). It is demonstrated that the proposed approach makes it possible to generate ultrasound images, the structure of which corresponds to the specified segmentation masks. It is expected that this method will improve the accuracy of solving the direct problem of automatic segmentation of the left ventricle.

The possibility of an ultrasound study of the heart (Echocardiography, EchoCG) is widely used in modern cardiology. This non-invasive technology allows studying cardiac activity of the patient by determining the global contractility of the heart muscle. The methods, which used in echocardiography, require performing manual operations from specially trained medical professionals. A number of researchers are working on the problem of automation of this medical technology. The article shows the way of solving the problem of the left ventricle area identification in echocardiography records with machine learning techniques. The task of the left ventricle delineation is reduced to the problem of pixels classification on video frames. A pixel can belong to one of two classes (the background region or the region of the left ventricle). The possibility of solving the task was tested with the following classifiers: decision tree, gaussian naive Bayes, linear discriminant analysis, quadratic discriminant analysis, adaboost classifier, random forest classifier. The assessment of classification results was performed using ROC curves. The best performance was obtained for both algorithms: decision tree classifier (AUC 0.93) and random forest classifier (AUC 0.93).

Article is devoted to the system development allowing to restore the volume of the left ventricle of heart, to estimate final systolic and diastolic volumes on the basis of the sequence of MRT-images from a parasternal position of a short axis in the automatic mode. The realized system was built on convolutional neural networks, 500 patients were used for training, for testing 200.