Сonvolutional neural networks in the diagnosis of skin neoplasms

• DOI: 10.26583/bit.2021.4.09
• Publication Date: 2021-12-07T07:30:27Z
• Authors: Valentin G. Nikitaev; Alexander N. Pron...; Olga B. Tamrazova; Vasily Yu. Sergeev; Yuri Yu. Sergeev; Dmitry V. Gurov; Sergei M. Zaitsev; Mikhail Solomatin...; Tamara P. Zanegina; Vladimir S. Vladi...
• Keywords: Contextual image classification; Melanoma diagnosis; Residual neural network

The problem of using artificial intelligence technologies in the diagnosis skin neoplasms is considered. Dermatoscopic images for 8 nosologies were considered as object research. melanoma was one them. Melanoma responsible most deaths all cancers. aim study to evaluate effectiveness use pre-trained convolutional neural networks classification neoplasms. A algorithm an ensemble proposed. Pre-trained have been studied form ensemble. Neural network samples selected from a set that proven themselves ImageNet Large Scale Visual Recognition Challenge. According results experiment best three eight inclusion – MobileNet_v2, ResNet_152, ResNeXt_101_32x8d. conducted on sample 10015 representing nosologies. average accuracy 79%. paper highlights features ensuring information security when telemedicine diagnostic proposed approach recognition work can be used design medical decision support systems malignant (including melanoma).