- Статьи
- Science and technology
- Sound application: AI will help determine the type of brain tumor by MRI
Sound application: AI will help determine the type of brain tumor by MRI
Russian scientists have developed a web application based on artificial intelligence that analyzes MRI scans of the brain and identifies tumors on them. Unlike analogues, in addition to the presence of pathology, the program determines its exact location and type. The accuracy of neoplasm detection is 97.1%. The development is open source and is available for free to everyone. According to experts, machine vision technologies are now in high demand among doctors all over the world and the market for such services will only increase.
An application for the diagnosis of brain tumors
Scientists at the I.M. Sechenov First Moscow State Medical University have developed a web application for the automatic detection of brain tumors on MRI images using computer vision techniques. The program detects the tumor, determines its exact location and classifies the neoplasm.
— The web application not only finds the tumor and determines its location, but also classifies the neoplasm, that is, makes a tentative conclusion about its nature. The model determines that the detected formation is, for example, glioma, meningioma or pituitary adenoma," said Ivan Simonovich, the author of the project, a graduate of the Master's degree in Intelligent Information Technologies in Medicine at the Advanced Engineering School of Sechenov University.
The web application created by Russian developers is open source. For their purposes, they trained one of the most modern open models of neural networks, YOLO v11. To do this, we used more than 5,000 standardized images collected from open data sets. Thanks to the use of the most up-to-date technologies and a series of experiments, it was possible to achieve an accuracy rate of 97.1% in tumor detection.
According to scientists, all similar free AI solutions for the automatic detection of tumors on MRI images are mainly limited to detecting the neoplasm without determining its location or using outdated models. And modern complex systems (Aidoc, Mediaire mdbrain, NeuroQuant Brain Tumor, and others) are usually commercial and use closed architectures and private databases. In addition, these solutions have been developed abroad and access to them in Russia is difficult.
The developers already have a prototype of the web application. They also created a server application with a web interface to demonstrate the operation of a trained model in real time. Further work will be aimed at improving the accuracy of the model, adding new data sources from Russian radiological laboratories, and finalizing the server application for the convenience of doctors. In the future, it is also planned to expand the functionality of the application — to train the model to find not only tumors, but also other brain pathologies on MRI scans.
— An AI solution for automatic detection of brain tumors in the future may help to increase the accuracy and speed of the primary diagnosis of these pathologies, as well as reduce the burden on radiologists and avoid the human factor in the process of interpreting images. The project is already ready for testing at the university clinics. It is gratifying that graduates of our master's and bachelor's degree programs demonstrate a high level of knowledge and skills and have all the competencies to create new products for practical healthcare," said Georgy Lebedev, the project's scientific director, director of the Center for Digital Medicine at Sechenov University.
Global market
Neuroradiology is one of the most complex areas of radiation imaging, and the service being developed by Russian scientists has good potential, says Alexander Kapishnikov, head of the Department of Radiation Diagnostics and Radiation Therapy at SamSMU.
— The main advantages of the project are the use of open source and the use of a powerful learning model for working with an array of MRI images. However, in order to achieve a real clinical application of the system, verification on real arrays of domestic digital data accumulating digital images at the regional and federal levels, as well as in-depth expert assessment, is necessary. In addition, the development should be brought into line with the state standard for artificial intelligence systems in medicine, which will be introduced in 2022," said Alexander Kapishnikov.
At the same time, it is important to remember that digital systems, including those using AI, can help in decision-making and ensure that a second opinion is obtained, which increases the accuracy of the diagnostic conclusion. But the final decision remains with the person, the expert emphasized.
The increase in the number of projects in Russian medicine using computer vision corresponds to the global trend. The global market for such technologies is estimated at $2 billion, and by 2030 its volume will reach about $15 billion, Vasily Korol, director of Data and Digital Technologies at AstraZeneca biopharmaceutical company, Russia and Eurasia, told Izvestia.
— One of the main drivers of growth will be the availability of large impersonal datasets for training AI models. It is the quality and volume of medical data that affects how accurately computer vision will work: for the diagnosis of complex, variable diseases, it is critically important to have a variety of data sets that cover both standard and atypical cases," he said.
According to the expert, for the further development of such projects in Russia, it is necessary to organize the process of exchanging anonymized medical data, as is done, for example, in China and the UAE. There are national exchanges of big data, which are available, among other things, to scientific organizations and innovative companies to create new tools in the healthcare system.
Переведено сервисом «Яндекс Переводчик»
