Celsus (Cels) Neural network for analysis of medical images

Main articles:

• Neural networks (neural networks)
• Artificial Intelligence in Medicine
• Artificial intelligence in radiology

Cels is artificial intelligence for analyzing medical images.

2021: Registration as a medical product for radiology in the areas of mammography and fluorography

On June 2, 2021, it became known that the system for analyzing medical radiographic images using artificial intelligence "Cels" (LLC "Medical Screening Systems") received a registration certificate for a medical product for radiology in the areas of mammography and fluorography (3 risk class in accordance with the order of the Ministry of Health of the Russian Federation dated July 7, 2020 No. 686n). The registration certificate confirms the quality, efficiency and security of the solution.

В Russia registered the first Cels medical decision-making assistance system using artificial intelligence technologies for use in screening breast cancer and chest organ pathologies

As explained, Cels, a medical decision-making assistance system (SPPVR) in radiology, is included in the register of domestic software. The platform recognizes the presence of benign or malignant changes in radiographic images, indicates their localization, and then interprets the results according to international standards.

According to the chief freelance specialist in radiation and instrumental diagnostics of the Moscow Department of Health and the Ministry of Health of the Russian Federation in the Central Federal District, director of the Center for Diagnostics and Telemedicine Sergey Morozov, Cels has already actually passed the phase of post-registration tests of medical artificial intelligence as part of the Moscow experiment on computer vision.

Before connecting to the common information circuit of ERIS EMIAS, the service passed validation on independent sets of studies in order to check the diagnostic accuracy, sensitivity and specificity of algorithms in the "field." Then, during 2020, Cels was used in real clinical practice by Moscow radiologists. This allowed developers to receive expert feedback from medical users and identify weaknesses for pre-training the service. told by Sergey Morozov

The use of Cels in practical work during screening will minimize the risks associated with the "human factor," compensate for the insufficiency of primary personnel, and increase the detection of cancer in the early stages.

Celsus became a solution thanks to which doctors really felt the benefits of modern technologies in the field of artificial intelligence both in reducing the time to make a diagnosis and in the form of an objective assistant who insures against errors. The region is already ready to use the service. said Alexander Korolev, Deputy Minister of Health of Kaluga Region

The effectiveness of SPPVR Cels was confirmed during pilot and commercial tests in a number of regions, as well as the experiment of the Moscow Department of Health on the use of artificial intelligence services in radiology.

Qualitative and accurate diagnosis at an early stage significantly increases the effectiveness of therapy and, as a result, the likelihood of a favorable prognosis for patients with cancer. That is why diagnostic solutions based on artificial intelligence are so in demand. The Cels system is one of the participants in the accelerator of startups iLab, organized by AstraZeneka together with the Skolkovo Foundation, within which we will be able to evaluate the optimal way to integrate the system into clinical practice. noted Evgenia Logacheva, Medical Director of AstraZeneka in Oncology

In 2020, the company attracted investments of the STI venture fund in the amount of 180 million rubles for the development of sales and technological improvement of the platform. As of June 2021, the solution is undergoing a procedure for assessing compliance with the requirements of the directives and harmonized standards of the European Union for entry into the international market.

2020

Search for radiologists to mark up digital images

The company "Medical Screening Systems," developing a system for supporting medical decision-making based on the technologies of AI "Cels," on December 2, 2020 announced the search for radiologists to mark up digital medical images. Marked data will be used for further training of artificial intelligence. Specialists who have passed the qualifying stages will have the opportunity to long-term cooperation with the company on a commercial basis.

Artificial Intelligence Celsus analyzes digital medical images and reveals the slightest signs of various pathologies, including oncology at an early stage. The use of the service is designed to reduce the burden on the doctor, reduce the time for analyzing studies and minimize the risks of missing pathology. At the moment, the system works in four areas of diagnosis: mammography, fluorography, computed tomography of the lungs and pathomorphology.

The target for December 2020 is used in 13 regions of the Russian Federation as part of pilot projects. Cels passed the stage of checking the stated accuracy on real verified data and as of December 2020 is the only service that has passed to the stage of industrial operation in the areas: mammography and fluorography.

Training and improving the work of artificial intelligence requires a large amount of verified data - and to prepare them requires markup performed by highly qualified radiologists. The marking process is the determination of objects of different classes on the image (malignant and benign neoplasms, calcinates, lymph nodes, areas of density, artifacts of images, etc.). Each snapshot is independently marked by several doctors, in case of discrepancies, the images are given for additional research.

The selection of radiologists will take place in two stages. The first is the classification of 50 non-trivial mammograms on the principle of "cancer/not cancer." At least 85% of snapshots must be correctly classified to complete this step. The second stage is the marking of various types of objects on mammograms in 4 projections. Objects are placed as "masks" in a specialized markup service.

According to the results of the first two stages, depending on the quality of the classification and the level of marking, radiologists receive an individual offer of cooperation with Cels.

Inclusion in the Register of Domestic Software

The first technology-based program artificial intelligence included in Register of domestic software was the Cels service developed the Russian by the company "" (Medical System Screening Order from Ministry of Digital Development, Communications and Mass Media Russia 31.08.2020 No. 429). The developer announced this on September 3, 2020.

Cels is used in mammography, fluorography, histology and computed tomography of the lungs. He analyzes digital medical images and reveals on them the slightest signs of pathological changes, including oncology in the early stages. The use of artificial intelligence reduces the time to conduct research and minimizes the risks of missing pathology.

The inclusion of artificial intelligence "Cels" in the register confirms its Russian origin and gives the development company the right to work with state institutions. As of September 2020, pilot projects for the use of the service have already been launched in 13 regions of the Russian Federation. Cels also participates in the Experiment of the Moscow Department of Health on the use of AI services, within the framework of which he has already processed more than 60 thousand studies on mammography and fluorography.

Development of the first version of the system for COVID-19 diagnostics

The Medical System Screening team, the developer of the Cels system, completed the development of the first version of the system for analyzing CT studies of lungs and detecting signs of coronavirus like SARS-COV-2 on them. This became known on July 6, 2020.

neuronets the Russian Foreign sets were also used for training. data As examples with absence COVID-19 , old CT studies without signs of viral pneumonia, made in 2019 and earlier, are also involved. According to epidemiologists, it will be possible to talk about the victory over COVID-19 only after the invention of the vaccine and passing clinical trials. And until that time, it is important to focus on the diagnosis and treatment of this disease.

The idea of ​ ​ developing a solution for coronavirus has been in the air for a long time. But we did not want to 'hype' and get to work until it is confirmed that CT scan is well suited for diagnosing this disease. We also waited for enough data to come up with decent quality. As soon as all the stars coincided, we immediately began work, "said Yevgeny Nikitin, technical director of Cels.

On July 6, 2020, a segmentator is being developed to highlight affected areas, work continues to improve the architecture and performance of the system. In the future, Cels will identify not only COVID-19, but also other diseases. The neural network is trained to find signs of oncology and various types of pneumonia on tomograms.

Celsus neural network will be able to determine the presence of tumor cells in breast tissues

The creators of the neural network for analyzing Cels medical images began work in a different direction - pathomorphology. Now the artificial intelligence of the service will be able to analyze histological studies: to study samples of breast tissues and determine the presence of tumor cells. This was announced on June 30, 2020 by Medical System Screening.

Making an oncological diagnosis is impossible without histological examination, it helps to determine the type of neoplasm and its characteristics. To perform such an analysis, you need to take a fragment of tissue from the patient for study by a pathomorphologist. Tissue samples are scanned, digitized, and instead of traditional microscopic examination are displayed. At this stage, artificial intelligence can help work: Cels will independently find areas with pathological cells and separate affected areas from healthy ones. In the future, it is planned to train the system to mark suspicious areas.

Scientific work is carried out jointly with the Nizhny Novgorod Regional Clinical Oncological Dispensary. Qualified doctors take part in the preparation of data and their markup - the developers have a database with verified cases of cancer.

Histological analysis is a complex and time-consuming study that is not available in all regions of the country. Tissue samples are often sent to the laboratories of the nearest large cities, which not only requires additional time, but also creates an additional burden on specialists. The help of artificial intelligence in histological research will reduce the burden on pathomorphologists and reduce the risks of incorrect and inaccurate diagnostics.