- Статьи
- Society
- Close the cell: in the Russian Federation created AI for predicting lung cancer metastases
Close the cell: in the Russian Federation created AI for predicting lung cancer metastases
Russian scientists have developed a neural network to detect cancer cells in lung tissue samples of adenocarcinoma. According to medics, if a tumor grows into blood vessels, from there it can spread further and form metastases. Therefore, it is important to identify areas of invasion before the tumor cells have entered the bloodstream - this allows you to predict the further development of the disease and adjust therapy. How AI will help speed up this process - in the material "Izvestia".
What is dangerous lymphovascular invasion
Scientists at the University of Sechenov have developed a neural network to detect lymphovascular invasion, that is, the penetration of cancer cells in the blood vessels, in samples of lung tissue in adenocarcinoma. This tumor develops in cells that secrete mucus and other bodily fluids. The development will make it possible to more accurately and quickly determine the risks of developing metastases and, if necessary, change the patient's therapy regimen.
With lymphovascular invasion, cancer cells are detected inside blood or lymphatic vessels. This is one of the unfavorable prognostic factors of the course of the disease, which requires in some cases the appointment of adjuvant, i.e. additional, therapy. However, the exact detection of invasion is difficult due to possible differences in the assessment of pathologists.
- If the tumor sprouts into the vessels, from there it can spread further and form metastases. Therefore, it is important to identify areas of invasion before tumor cells have entered the bloodstream. This allows you to predict the further development of the disease and adjust therapy. However, the interpretation of histologic data may differ from specialist to specialist, and there is a risk of not noticing invasion in time. Our research is aimed at creating a system for automatic assessment of invasion, which will help pathologists to obtain more objective and accurate results," Alexey Faizullin, head of the digital microscopic analysis laboratory, explained to Izvestia.
The core of the system was a neural network, for training which the scientists used 162 histoscans containing 8212 vessels marked by pathologists. The AI was able to identify blood vessels and tumor invasion in them with an accuracy of more than 95%.
AI in oncomedicine
The scientists also tested whether their approach would enable faster identification of regions of invasion in a pilot experiment. As it turned out, the use of the system can reduce the analysis time by 17% on average, and in particularly complex cases - by 20%.
In addition to clinical practice, the neural network may be useful in science. Simpler and faster processing of histoscans will make it possible to better study the peculiarities of lymphovascular invasion and identify new aspects of its influence on the development of metastases and tumor biology in general.
The scientists plan to further work on improving the accuracy and efficiency of the neural network, training it to work with vessels of other organs, comparing lymphovascular invasion with other prognostic factors, such as the proliferation index of tumor cells, as well as creating a multimodal prognostic model.
In the long term, they expect to create a fully automated service that can be integrated into medical information systems.
Right now, this system detects tumor invasion in blood vessels with an accuracy of over 95%. This is a very strong result, because the use of AI in this case avoids the human factor and accelerates tumor analysis by as much as 20%. Given the shortage of specialists in this field and the ever-growing demand for these operations, this will seriously speed up the work of doctors and specialists in the field, and help people to receive the required treatment faster, said Anton Averyanov, CEO of ST IT Group, an expert of the TechNet NTI market.
- It is expected that this system will always be used and will speed up the work of identifying them. Over time, the accuracy will increase, because now the neural network is trained only on 162 histoscans. If we bring this number up to 1,000 or more, the accuracy may approach 98 or even 99%, which will allow a greater reliance on the system and which will speed up diagnosis even more," the expert said.
This could be a fundamentally useful approach when it reaches implementation, said Yury Molodykh, director for development of technological contests of NTI Up Great Foundation.
- In medicine, AI is just beginning to be used. Due to the great responsibility, doctors are not ready to entrust key tasks, on which a patient's life may depend, to AI if they do not realize that it can be trusted. Therefore, the introduction will be smooth and gradual to make sure that it will be safe for patients," the expert emphasized.
The work was conducted together with colleagues from Yaroslav the Wise Novgorod State University within the framework of the program of creation and development of the NCMU "Digital Biodesign and Personalized Healthcare". The results of the study were published in the Journal of Pathology Informatics.
