Abstract. In modern medicine, the analysis of texture features of images plays an important role in the diagnosis of various diseases. This study is devoted to the methods of classifying texture features of burn injury images, which allow identifying informative characteristics and increasing the accuracy of diagnosis. Modern machine learning algorithms used for automated image analysis are considered. A comparative analysis of various classification methods is carried out, the most effective approaches to processing texture features are identified. The results of the study can be used to develop intelligent systems to support medical decision-making. The study used 1,500 burn images (500 for each class) obtained from open sources Roboflow, Kaggle. Texture features extracted using MaZda software were used for further classification. Initially, 279 features were extracted, from which 21 most significant features were selected using 6 selection methods: ANOVA, Fisher, Relief, SBS, SFS and RFE. A review of current approaches to automated burns analysis, including image processing and machine learning methods, published in the last five years is performed. The work demonstrates the promising application of machine learning in medicine and the need for further research to improve classification accuracy and practical implementation of the developed algorithms.

Keywords: texture features, MaZda, automated burn analysis, medical diagnosis, image processing, burn classification.