E.V. Romanova1, V.A. Barteva2, S.D. Kim3

1–3 Financial University under the Government of the Russian Federation (Moscow, Russia)
1 EkVRomanova@fa.ru, 2 VABarteva@fa.ru, 3 215922@edu.fa.ru

This article compares modern deep learning methods for dermatoscopy image classification and, based on the experimental data obtained, substantiates the choice of an architecture optimal for practical implementation in a clinical environment. With the increasing use of data processing technologies and the capabilities of computer systems, deep learning methods are becoming increasingly po­pular for solving image classification problems. Convolutional neural networks (CNNs), in particular, demonstrate high performance in various fields – from medicine to industry and biometrics – where the accuracy and speed of visual information analysis are crucial. Particular attention is paid to the analysis of dermatoscopy images, which is crucial for the early diagnosis of skin diseases, including melanoma, one of the most dangerous malignant tumors in humans. The use of convolutional neural networks in this context allows for the identification of complex characteristics and consideration of the specifics of this highly detailed data, which includes multiple ambiguous criteria. Furthermore, the need to accurately select hyperparameters and choose the optimal layer structure and optimization methods requires significant computational resources and time. In recent years, numerous CNN architectures have emerged, such as EfficientNet, DenseNet, Inception, and ResNet, improving the performance of neural networks. However, adapting them to specific classification tasks remains a challenge. This study aims to compare various deep learning approaches for dermatoscopy
image classification. This will help identify the most effective methods for practical use in medical diagnostics.

Romanova E.V., Barteva V.A., Kim S.D. A study on the effectiveness of deep learning methods in the task of dermatoscopy image classification. Dynamics of complex systems. 2026. V. 20. № 3. P. 82−90. DOI: 10.18127/j19997493-202603-09 (in Russian).

1. Cheng Yu., Anikeev E.A. Thoughts on convolution kernel of convolutional neural network. Novy`e aspekty` modelirovaniya sistem i processov. Otv. red. V.K. Zol`nikov, A.I. Zarevich. Materialy` Mezhdunar. nauchno-prakt. konf. (Voronezh, 26 maya 2023 g.). Voronezh: Voronezhskij gosudarstvenny`j lesotexnicheskij universitet im. G.F. Morozova. 2023. P. 483–491. DOI 10.58168/NAMSP_483-491
2. Bagaev I.I. Analiz ponyatij nejronnaya set` i svertochnaya nejronnaya set`, obuchenie svertochnoj nejroseti pri pomoshhi modulya TensorFlow. Matematicheskoe i programmnoe obespechenie sistem v promy`shlennoj i social`noj sferax. 2020. T. 8. № 1. S. 15–22. DOI 10.18503/2306-2053-2020-8-1-15-22
3. Cheng Yu., Anikeev E.A. Understanding convolutional neural networks. Novy`e aspekty` modelirovaniya sistem i processov. Otv. red. V.K. Zol`nikov, A.I. Zarevich. Materialy` Mezhdunar. nauchno-prakt. konf. (Voronezh, 26 maya 2023 g.). Voronezh: Voronezhskij gosudarstvenny`j lesotexnicheskij universitet im. G.F. Morozova. 2023. P. 230–236. DOI 10.58168/NAMSP_230-236
4. Narkevich A.N., Vinogradov K.A., Paraskevopulo K.M., Mamedov T.X. Intellektual`ny`e metody` analiza danny`x v biomedicinskix issledovaniyax: svertochny`e nejronny`e seti. E`kologiya cheloveka. 2021. № 5. S. 53–64. DOI 10.33396/1728-0869-2021-5-53-64
5. Chudakov G.A., Dzhunkovskij A.V. Diagnostika melanomy` putem analiza dermatoskopicheskogo izobrazheniya pri pomoshhi svertochnoj nejronnoj seti. Mezhdunarodny`j zhurnal prikladny`x i fundamental`ny`x issledovanij. 2022. № 9. S. 63–69. DOI 10.17513/mjpfi.13442
6. Mixajlusov A.V. Ispol`zovanie svertochny`x nejronny`x setej dlya diagnostiki raka kozhi. Vestnik Voronezhskogo gosudarstvennogo texnicheskogo universiteta. 2023. T. 19. № 1. S. 20–26. DOI 10.36622/VSTU.2023.19.1.003
7. Xu T., Zhou Yi., Ji K., Liang Y. When will gradient methods converge to max‐margin classifier under ReLU models. Stat. 2021. V. 10. № 1. P. 354. DOI 10.1002/sta4.354
8. Micikevicius P. et al. Mixed Precision Training. ICLR. 2018.
9. Prokopchina S.V., Zvyagin L.S. Aktual`ny`e napravleniya i perspektivy` razvitiya prikladny`x informacionno-izmeritel`ny`x sistem dlya resheniya zadach avtomatizacii monitoringa i upravleniya v slozhny`x texnicheskix kompleksax. Dinamika slozhny`x sistem – XXI vek. 2025. T. 19. № 5. S. 110–125. DOI: 10.18127/j19997493-202505-13 (in Russian).