Skolkovo resident, Care Mentor AI, uses Zhores supercomputer to identify the severity of COVID-19
Credit: Care Mentor AI
As part of the anti-COVID-19 program of the Skoltech Center for Computational and Data-Intensive Science and Engineering (CDISE), Care Mentor AI data scientists used Skoltech’s Zhores supercomputer to enhance pathology detection accuracy. Care Mentor AI is a Russian provider of neural network (NN) based computer vision services for the analysis of X-ray and computer tomography (CT) images.
“Thanks to Zhores, we performed a wealth of experiments and built a high-performance NN for our system that estimates the severity of COVID-19 based on full 3D chest CT scans. I am confident that our continuing collaboration will help improve Care Mentor AI’s computer vision services to reduce the workload on radiologists and make patient diagnostics faster,” says Pavel Roitberg, a co-founder of Care Mentor AI.
Care Mentor AI’s AI-based system analyzes and interprets the radiological imaging data using a very fast and accurate NN-based technology that helps identify malignant tumors, tuberculosis, pneumonia, and other health problems. The system estimates the severity of COVID-19 and the extent of damage it caused in percent based on both radiological images and CT scans. CT is more accurate when it comes to detecting symptoms of viral lung damage and, primarily, interstitial pulmonary infiltration, which is hard to discern in an X-ray image. It is only with CT that one can get a true picture of the localization, spread, and progression of the disease, which is extremely important for assessing its severity and behavior.
The first energy-efficient petaflops supercomputer in Russia, Zhores is designed specifically for machine learning (ML) and data-based modeling tasks and aims to help Skoltech’s scientists and its academic and industrial partners in making breakthroughs in medicine and other fields.
“We built Zhores around an architecture that can handle large amounts of data, especially biomedical research data, which eventually helped us successfully train Care Mentor AI’s neural networks that are used to analyze health problems revealed by radiological images. Working together and building on our partner’s research capabilities and achievements, we will be able to help Russian medicine make a quantum leap in technology. Biomedical data has a very high dimension and, therefore, requires parallel computing. Simply put, we are quite ahead of time with this research thanks to Zhores’s architecture,” comments Maxim Fedorov, Skoltech Vice President for Artificial Intelligence and Mathematical Modelling.
Care Mentor AI experts report that they have trained their neural networks to detect cancers in CT images with 95% accuracy, estimate the percentage of lung damage caused by COVID-19 with 86% accuracy, and calculate the foot arch angle to diagnose flat feet with 99% accuracy. Moreover, the Care Mentor AI application marks and prioritizes abnormalities detected in radiological images with 93% accuracy, thus helping doctors to cope with a larger number of cases.
Pilot testing of Care Mentor AI’s chest X-ray screening system is successfully underway at PJSC Medicina a Moscow clinic founded by Academician Grigory Roitberg. The application is also being tested within the Moscow Experiment, a program that aims to test and integrate innovative computer-vision-based medical image analysis technologies in the Moscow healthcare system. In the course of the Experiment, Care Mentor AI’s service was integrated into over 40 medical institutions in Moscow.
Care Mentor AI’s CT analysis system capable of estimating the severity and extent of damage caused by COVID-19 has been successfully operating at the Ivanovo Region reference center which collects CT images with suspected COVID-19 from hospitals across the Region.