K.M. Paraskevopulo¹, A.N. Narkevich², K.A. Vinogradov³

1,2 FSBEI HE Prof. V.F. Voino-Yasenetsky KrasSMU MOH Russia (Krasnoyarsk, Russia); 

3 Department of Medical Faculty, FSBEI HE Prof. V.F. Voino-Yasenetsky KrasSMU MOH Russia (Krasnoyarsk, Russia)   

Dermatoscopy is non-invasive method of skin cancer revealing. This method took a strong position in Russia in the 20th century. The next step in the evolution of dermatoscopy can be development of decision support systems for doctors, which do not have sufficient experience in diagnosis of skin malignant neoplasms. For development of such systems, it is necessary to take artificial neural networks and telemedicine technologies as a basis. Artificial neural networks can help to reveal skin malignant neoplasms on digital skin images. Telemedicine technologies are needed to transfer the results obtained to specialized centers in order to further analysis of the results by the doctor and to call the patient for a consultation. The aim of this study is development and convolution neural networks work result analysis. The study used 7088 digital images of skin with histologically confirmed absence or presence of malignant neoplasms from The International Skin Image Collaboration (ISIC) database, of which 4881 images are benign and 2207 are malignant. For all used convolutional neural networks (MobileNetV2, MobileNet, DenseNet121, DenseNet169 и DenseNet201, Xception, VGG16 и VGG19, ResNet50, ResNet101 и ResNet152, ResNet50V2, ResNet101V2 и ResNet152V2, InceptionV3, InceptionResNetV2, NASNetMobile), we made fully connected layer with two neurons, which are classified digital skin images on malignant and benign skin neoplasms. The Softmax function is used as the activation function for two output neurons. We used accuracy for quality control of convolutional neural networks malignant and benign skin neoplasms classification in train, validation and test sample. For these indicators, a 95% confidence interval was calculated. ResNet50 convolutional neural network showed highest value of accuracy on test sample – 83,61% [83,15; 84,03]. Training was made with preloaded weights from training on image-net images. Also we used additional training on digital skin images. ResNet50 neural network architecture contains 50 convolutional layers, which distinguishes it from the ResNet101 and ResNet152 models, which contain 101 and 152 layers, respectively. Increasing the number of convolutional layers reduces the classification accuracy of malignant and benign neoplasms. In results of development and convolutional neural networks work analysis, it was found that convolutional neural network built by type ResNet50 can be used in development of decision systems for doctors for malignant skin neoplasms diagnostic on digital skin images in future.

Paraskevopulo K.M., Narkevich A.N., Vinogradov K.A. Application of convolutional neural networks for malignant skin lesions recognition on digital skin images. Technologies of Living Systems. 2021. V. 18. № 2. Р. 31–38. DOI: https://doi.org/10.18127/j20700997-20210204 (in Russian).

1. Alekseyeva O.V., Rossiyev D.A., Ilyenkova N.A. Primeneniye iskusstvennykh neyronnykh setey v differentsialnoy diagnostike retsidiviruyushchego bronkhita u detey. Sibirskoye meditsinskoye obozreniye. 2010. T. 66. № 6. S. 75–79 (in Russian).
2. Sergeyev Yu.Yu., Sergeyev V.Yu. Primeneniye dermatoskopii v prakticheskoy dermatologii. Kremlevskaya meditsina. Klassicheskiy vestnik. 2018. № 1. S. 8–15 (in Russian).
3. Sergeyev Yu.Yu., Olisova O.Yu. Analiz obrashchayemosti na dermatoskopicheskiy osmotr. Rossiyskiy zhurnal kozhnykh i venericheskikh bolezney. 2016. T. 19. № 2. S. 107 (in Russian).
4. Tarasenko G.N., Kukushkina S.V., Tarasenko Yu.G. Dermatoskopiya: metod neinvazivnoy diagnostiki v praktike dermatologa. Gospitalnaya meditsina: Nauka i praktika. 2020. T. 1. № 1. S. 13–16 (in Russian).
5. Bandic J., Kovacevic S., Karabeg R., Lazarov A., Opric D. Teledermoscopy for Skin Cancer Prevention: a Comparative Study of Clinical and Teledermoscopic Diagnosis. Acta informatica medica. 2020. V. 28. № 1. P. 37–41.
6. Brinker T.J., Hekler A., Enk A.H., Berking C., Haferkamp S., Hauschild A., Weichenthal M., Klode J., Schadendorf D., Holland-Letz T., Kalle C., Frohling S., Utikal J.S. Deep neural networks are superior to dermatologists in melanoma image classification. European journal of cancer. 2019. № 119. P. 11–17.
7. Chollet F.Xception: Deep Learning with Depthwise Separable Convolutions. Computer Vision and Pattern Recognition. 2016.
8. He K., Zhang X., Ren S., Sun J.Deep Residual Learning for Image Recognition. Computer Vision and Pattern Recognition. 2015.
9. He K., Zhang X., Ren S., Sun J.Identity Mappings in Deep Residual Networks. Computer Vision and Pattern Recognition. 2016.
10. Howard A.G., Zhu M., Chen B., Kalenichenko D., Wang W., Weyand W., Andreetto M., Adam H. MobileNets: Efficient Convolutional Neural Networks for Mobile Vision Applications. Computer Vision and Pattern Recognition. 2017.
11. Huang G., Liu Z., Maaten L., Weinberger K.Q. Densely Connected Convolutional Networks. Computer Vision and Pattern Recognition. 2016.
12. Sandler M., Howard A., Zhu M., Zhmoginov A., Chen L. MobileNetV2: Inverted Residuals and Linear Bottlenecks. The IEEE Conference on Computer Vision and Pattern Recognition (CVPR). 2018.
13. Sirmonyan K., Zisserman A.Very Deep Convolutional Networks for Large-Scale Image Recognition. Computer Vision and Pattern Recognition. 2014.
14. Sivaraj S., Malmathanraj R., Palanisamy P. Detecting anomalous growth of skin lesion using threshold-based segmentation algorithm and Fuzzy K-Nearest Neighbor classifier. Journal of Cancer Research and Therapeutics. 2020. V. 16. № 1. P. 40–52.
15. Szegedy C., Vanhoucke V., Ioffe S., Shlens J., Wojna Z.Rethinking the Inception Architecture for Computer Vision. Computer Vision and Pattern Recognition. 2015.
16. Szegedy C., Ioffe S., Vanhoucke V., Alemi A.Inception-v4, Inception-ResNet and the Impact of Residual Connections on Learning. Computer Vision and Pattern Recognition. 2016.  
17. Zoph B., Vasudevan V., Shlens J., Le Q.V.Learning Transferable Architectures for Scalable Image Recognition. Computer Vision and Pattern Recognition. 2015.