As part of the program to fight COVID-19 at the Skoltech Center for Scientific and Engineering Computing Technologies for Large Data Problems (CDISE), data scientists from Care Mentor AI were able to use the Zhores supercomputer to improve the accuracy of pathology detection.
“Thanks to the use of the Zhores supercomputer, we were able to carry out many experiments in a short time and obtain a high-quality neural network for our service for determining the severity of COVID-19 on complete 3D CT examinations of the chest. I am confident that our further cooperation will lead to the improvement of computer vision services Care Mentor AI, thereby reducing the workload on radiologists and allowing faster diagnostics of patients, "says Pavel Roitberg, co-founder of Care Mentor AI, a Russian developer of computer vision services for research analysis. in the field of medical image diagnostics based on neural network algorithms.
Care Mentor AI product is an artificial intelligence system for analyzing and interpreting the results of medical image diagnostics. Neural network technology analyzes the results of medical image diagnostics and, with high speed and accuracy, determines the presence or absence of pathologies in patients, including malignant neoplasms, tuberculosis, pneumonia, etc. COVID-19. Computed tomography makes it possible to more accurately detect pathological signs inherent in viral lung damage, first of all, interstitial infiltration of lung tissue, which is rather difficult to do on an X-ray. In addition, only CT allows you to objectively determine the localization and prevalence of the pathological process and the transformation of one stage of the course of the disease into another, which is extremely important for assessing the severity and dynamics of the pathological process.
Zhores is the first petaflops energy efficient supercomputer in Russia, specially designed for solving problems of machine learning and data-based modeling. This state-of-the-art computing system aims to help Skoltech scientists and its academic and industrial partners make breakthroughs in medicine and other fields.
“In developing the Zhores supercomputer, we laid down an architecture that meets the requirements for working with large amounts of data, especially in the field of working with biomedical research data. This is what allowed us to successfully train the neural networks of the Care Mentor AI company, which are used to analyze pathologies in radiation diagnostics studies. Strengthening the capabilities and developments of our partner, together we will be able to bring medicine in Russia to a new technological level. Biomedical data has a huge dimension, therefore, to work with it, computing power is required with the ability to parallelize all processes. Thus, our developments were somewhat ahead of their time thanks to the architecture of the Zhores supercomputer, ”comments Maxim Fedorov, Skoltech Vice President for Artificial Intelligence and Mathematical Modeling.
Specialists of the Skolkovo project member Care Mentor AI report that the company has already taught neural networks to “see” cancer foci on CT scans with an accuracy of 95%, to calculate the percentage of lung damage in COVID-19 infection with an accuracy of 86%, and to calculate the angle of inclination of the foot. and establish the presence or absence of flat feet with an accuracy of 99%. Moreover, the Care Mentor AI service is able to prioritize and mark pathologies on X-ray examinations with an accuracy of 93%, helping the doctor to cope with the volume of work faster and have time to carry out more diagnostic examinations.
The system for X-ray screening of pathologies of the chest organs is successfully undergoing pilot testing at JSC "Medicine" - the clinic of Academician Roitberg. Also, a product for X-ray screening of pathologies of the chest organs is being tested as part of the Moscow experiment on the use of innovative technologies in the field of computer vision for the analysis of medical images and further use in the health care system of the city of Moscow. In the course of this experiment, the service was integrated in more than 45 medical institutions in Moscow.