Neural network classifier for hyperspectral images of skin pathologies

• Published in: Journal of physics. Conference series Vol. 2127; no. 1; pp. 12026 - 12029
• Main Authors: Vinokurov, V, Khristoforova, Yu, Myakinin, O, Bratchenko, I, Moryatov, A, Machikhin, A, Zakharov, V
• Format: Journal Article
• Language: English
• Published: IOP Publishing 01.11.2021
• ISSN: 1742-6588; 1742-6596
• DOI: 10.1088/1742-6596/2127/1/012026

Abstract This paper describes the use and results of a neural network classifier trained on a set of hyperspectral images of benign and malignant neoplasms. The analysis is carried out on 2D images extruded from hyperspectral data. The ranges of wavelengths at which the research is carried out is represented by the intervals 530–570 nm and 600–606 nm, which is caused by the assumption that the analysis of the entire spectral range is redundant and the prospect of saving resources. Melanoma, basal cell carcinoma (BCC), nevus and papilloma are accepted as primary classes, as the most dangerous, most common and non-malignant types of neoplasms, respectively. The neural network classifier is based on a three-block VGG network. With a training set included 1944 images, the classification accuracy for 4 types of samples was 92%.