RUS  ENG
Full version
JOURNALS // News of the Kabardino-Balkarian Scientific Center of the Russian Academy of Sciences // Archive

News of the Kabardino-Balkarian Scientific Center of the Russian Academy of Sciences, 2025 Volume 27, Issue 2, Pages 103–112 (Mi izkab939)

Computer science and information processes

Models and methods of deep learning in medical image recognition and classification tasks

I. A. Pshenokovaab, M. R. Kiyasova

a Kabardino-Balkarian State University named after Kh.M. Berbekov, 360004, Russia, Nalchik, 173 Chernyshevsky street
b Institute of Computer Science and Problems of Regional Management – branch of Kabardino-Balkarian Scientific Center of the Russian Academy of Sciences, 360000, Russia, Nalchik, 37-a I. Armand street

Abstract: The paper presents a study and analysis of deep learning models and methods in the problems of recognition and classification of brain tumor images. To compare the effectiveness of the most relevant and available models based on convolutional neural networks, the VGG19, Xception, © Ïøåíîêîâà È. À., Êèÿñîâ Ì. Ð., 2025 INFORMATICS AND INFORMATION PROCESSES 104 News of the Kabardino-Balkarian Scientific Center of RAS Vol. 27 No. 2 2025 and ResNet152 models were selected. The Xception model showed the best results. The purpose of this work is to optimize and train the selected model using various methods to improve the accuracy of diagnosing human brain tumors. A strategy for improving this model using transfer learning and data augmentation methods is proposed and implemented. The tests show that the improved model demonstrates higher accuracy and resistance to various types of data distortions, which makes it more effective for image recognition and classification tasks.

Keywords: image recognition methods, deep learning methods, convolutional neural networks, transfer learning methods

UDC: 004.853

MSC: 68T42

Received: 14.03.2025
Revised: 07.04.2025
Accepted: 09.04.2025

DOI: 10.35330/1991-6639-2025-27-2-103-112



Bibliographic databases:


© Steklov Math. Inst. of RAS, 2025