Authors

Machine learning for detection of aortic root landmarks

A significant increase in the number of transcatheter aortic valve replacements entails the development of auxiliary systems that solve the problem of intra- or preoperative assistance to such interventions. The main concept of such systems is the concept of computerized automatic anatomical recognition of the main landmarks that are key to the procedure. In the case of transcatheter prosthetics – elements of the aortic root and delivery system. This work is aimed at demonstrating the potential of using machine learning methods, the modern architecture of the ResNet V2 convolutional neural network, for the task of intraoperative real-time tracking of the main anatomical landmarks during transcatheter aortic valve replacement. The basis for training the chosen architecture of the neural network was the clinical graphical data of 5 patients who received transcatheter aortic valve replacement using commercial CoreValve systems (Medtronic Inc., USA). The intraoperative aortographs obtained during such an intervention with visualization of the main anatomical landmarks: elements of the fibrous ring of the aortic valve, sinotubular articulation and elements of the delivery system, became the output data for the work of the selected neural network. The total number of images was 2000, which were randomly distributed into two subsamples: 1400 images for training; 600 – for validation. It is shown that the used architecture of the neural network is capable of performing detection with an accuracy of 95-96% in terms of the metrics of the classification and localization components, however, to a large extent does not meet the performance requirements (processing speed): the processing time for one aortography frame was 0.097 sec. The results obtained determine the further direction of development of automatic anatomical recognition of the main landmarks in transcatheter aortic valve replacement from the standpoint of creating an assisting system – reducing the time of analysis of each frame due to the optimization methods described in the literature. Read more...

Neural network analysis of coronary artery stenoses: assessment of the accuracy and speed of promising architectures

Significant interest in the field of application of machine learning for the analysis of medical images stimulates the search for promising algorithms for solving routine diagnostic problems in cardiology. In relation to cardiovascular diseases, such a procedure is coronary angiography, which assesses the state of the vascular network and the presence of stenotic areas. This paper demonstrates an example of using modern models of neural networks: SSD MobileNet V2, SSD ResNet-50, Faster-RCNN Inception ResNet for localizing a single-vessel coronary artery lesion on a set of clinical data (3200 images). It is shown that the Faster-RCNN Inception ResNet V2 model was the most accurate in terms of the chosen metric mAP[0.5:0.95], reaching 0.9434 and 0.95 for the validation and test sets, respectively. However, the data processing speed was 0.363 seconds per frame, which corresponds to a speed of 2.8 frames/sec, which does not correspond to the speed of coronary angiography (15 frames/sec). Neural networks with a more “simple” architecture demonstrated an unsatisfactory quality of stenosis localization, expressed in a low characteristic mAP[0.5:0.95]. The results of this study demonstrate a key problem in the application of machine learning algorithms on graphic data – high accuracy, which may be acceptable for medical diagnostic procedures, is “decompensated” by long-term image analysis, as a result, the use of unmodified neural network architectures does not provide real-time data processing. Read more...