According to the Russian Federation Ministry of Health, the mortality rate from cancer in Russia is currently more than 200 people per 100 thousand. Brain tumors are found in 65–80% of those who die from oncology. Therefore, diagnosing and monitoring the dynamics of neuro-oncological diseases in patients with primary and secondary brain tumors is extremely important. Early detection of neoplasms significantly improves the effectiveness of targeted therapy (surgical, radiation, chemotherapy, and immunological).
For his final thesis, Maxim Kochanov, a student at the NSU Mechanics and Mathematics Department, considered two problems. The first was semantic segmentation of brain tumors on MRI images and the second predicting the time interval of patients’ lives based on images of segmented brain tumors established solving the first problem. Academic Supervisor Bair Tuchinov, Head of the Laboratory for Streaming Data Analytics and Machine Learning at the Department, was his academic supervisor. To solve these problems, both machine learning and deep learning approaches were used. Data for training machine learning algorithms and neural networks was provided by the international competition Brain Tumor Segmentation Challenge – 2020.
Kochanov described the work. In the course of our thesis work, we developed a prototype for a personalized diagnostics system. An MRI image of the brain is fed to the system and as an output, we receive an image of a specific brain tumor and / or cerebral edema. If there is a tumor on the MRI image and / or swelling, we get the most likely time interval for the patient’s life. This means there is a high applied value in medicine for my thesis.
Previously, Novosibirsk scientists developed a technology to improve the diagnosis of brain tumors using artificial intelligence.