Vologda students trained a neural network to recognize lung cancer

A team of VoSU students – Pavel Smirnov, Elizaveta Shuvalova, Sergey Smekalov, under the guidance of Associate Professor of the Department of Automation and Computer Engineering Georgy Rapakov, worked to train a neural network to recognize early-stage malignant neoplasms in images, as well as to classify lung cancer.

The relevance of the problem is confirmed by statistical data: in the overall structure of oncology morbidity in the Vologda region, lung cancer makes up 10.1% of the total number of identified malignant neoplasms (MNT). This type of oncology also accounts for the largest share in the mortality rate of the population of the Vologda region from cancer – 17.4%, which is higher than the average for Russia.

“Such mortality statistics are due to the fact that upon initial treatment, almost 40% of people are diagnosed with cancer at the fourth stage. The goal of our development is to diagnose the disease at the first stage, since in this case the survival rate will be more than 90%,” — explained the VoSU student Elizaveta Shuvalova.

The lung cancer recognition method that Vologda students rely on is based on traditional image processing with a neural network.

“Specialized Internet sources made it possible to collect a database of 22 thousand images classified in accordance with the requirements of the International Classification of Diseases. Of these, 17 thousand images were used for training the neural network, and 5 thousand for testing. As a result, the neural network will be able to recognize different types of malignant neoplasms and distinguish healthy lungs from those affected by the disease,” – explains VoSU student Sergey Smekalov.

“The classification accuracy on the test sample ranged from 67 to 95%, which is considered an acceptable value for a preliminary diagnosis. The final decision always remains with the specialist and may require additional diagnostics. The most important thing here is not to waste time on research and start treatment as early as possible,” — says VoSU student Pavel Smirnov.

In the future, students plan to achieve even more accurate results by expanding the data set and using new artificial intelligence methods.

“Recognition of malignant neoplasms at an early stage, including lung cancer, is a topical topic for the healthcare system. Students of Vologda State University are going to propose an initiative development to the Department of Health of the Vologda Region. I hope our project will find a response from specialists, and together we will be able to discuss the possibilities of using a neural network approach and artificial intelligence methods in the work of healthcare institutions in the city and region, – noted Georgy Rapakov.