Authors

On anomalies detection in electrocardiograms with unsupervised deep learning methods

Anomaly detection is an important task in various applications and areas of technology and production, such as structural defects, malicious intrusions into management and control systems, financial supervision and risk management, digital health screening, etc. The ever-increasing flows of diverse data and their structural complexity require the development of advanced approaches to their solution. In recent years, deep learning methods have achieved significant success in detecting anomalies, and unsupervised deep learning methods have become especially popular. Methods of anomaly detection by methods of deep learning without a teacher are investigated in the work on the example of a set of electrocardiograms containing normal ECG signals and ECG signals of people with various cardiovascular diseases (anomalies). To detect abnormal electrocardiograms, an autoencoder model has been developed in the form of a deep neural network with several fully connected layers. Also, to solve this problem, a method is proposed for selecting the threshold for separating abnormal ECG signals from normal ones, consisting in optimizing the ratio of performance indicators of the autoencoder model by methods. The paper presents a comparative analysis of the effectiveness of applying various machine learning models, such as the one class Support Vector Method, Isolation Forest, Random Forest and the presented autoencoder model to solving the problem of detecting abnormal ECG signals. For this purpose, metrics such as accuracy, recall, completeness, and f-score were used. His results showed that the proposed model surpassed the other models in solving the problem with accuracy = 98.8% precision = 95.75%, recall = 99.12%, f1-score = 98.75%. Read more...

On segmentation of brain tumors by MRI images with deep learning methods

Segmentation of a brain tumor is one of the most difficult tasks in the analysis of medical images. The purpose of brain tumor segmentation is to create an accurate outline of brain tumor areas. Gliomas are the most common type of brain tumors. Diagnosis of patients with this disease is based on the analysis of the results of magnetic resonance imaging and segmentation of the tumor boundaries manually. However, due to the time-consuming nature of the manual segmentation process and errors, there is a need for a fast and reliable automatic segmentation algorithm. In recent years, deep learning methods have shown promising effectiveness in solving various computer vision problems, such as image classification, object detection and semantic segmentation. A number of methods based on deep learning have been applied to segmentation of brain tumors, and promising results have been achieved. The article proposes a hybrid method for solving the problem of segmentation of brain tumors based on its MRI images based on the U-Net architecture, the encoder of which uses a model of a deep convolutional neural network pre-trained on a set of ImageNet images. Among such models were used VGG16, VGG19, MobileNetV2, Inception, ResNet50, EfficientNetb7, InceptionResnetV2, DenseNet201, DenseNet121. Based on the hybrid method, the TL-U-Net model was implemented, and numerical experiments were carried out to train it with different encoder models for segmentation of brain tumors based on its MRI images. Computer experiments on a set of MRI images of the brain showed the effectiveness of the proposed approach, the best encoder model turned out to be the neural network Densenet121, which provided indicators of segmentation accuracy MeanIoU=90.34%, MeanDice=94.33%, accuracy=94.17%. The obtained estimates of segmentation accuracy are comparable or exceed similar estimates obtained by other researchers. Read more...

Heart coronary arteries detection by deep learning methods

Coronary artery disease is the most dangerous heart disease caused by coronary artery disease. In clinical practice, X-ray coronary angiography is the main imaging method used to diagnose coronary artery disease. High cost and complexity of analysis of a large amount of data by a cardiac surgeon made it necessary to automate the process of image processing and diagnosis of stenoses. In this paper we considered models of deep detection, localisation and characterisation of stenoses using popular models SSD, R-FCN, Faster-­RCNN, RetinaNet, EfficientDet. The models were pre-trained on the COCO image set and varied on the underlying neural network architecture. Computational experiments on stenosis detection from X-ray images were performed on the coronary angiography data used. The data consist of 9378 clinically acquired video sequences from invasive coronary angiography performed in DICOM format and labelled into individual frames for each video containing coronary artery stenosis. A total of 1593 image sequences with a resolution of (512×512) pixels were annotated. A comparative analysis of the models in terms of the main performance indicators: mAP accuracy, image processing time, number of model parameters was carried out. The obtained results allow us to state that the Faster R-CNN (ResNet101) and EfficientDet D4 (ResNet101) models are the detectors of choice in the detection of coronary artery stenosis. They have high detection accuracy and image processing speed compared to other models, as well as relatively low weights of parametrics. Comparative analysis of their performance with the results of other researchers showed superior or comparable results obtained in this work. Read more...