The article explores the problem of comparing and selecting radiomic and deep convolutional features extracted from CT images to enhance the accuracy of texture classification in CT diagnostics. By using the mRMR method, the study assesses the significance of these features in predicting genetic mutations in patients with lung cancer, highlighting their importance for refining diagnostic procedures. The developed predictive model demonstrates high classification accuracy of 92%, which indicates its high efficiency. Analysis of the results reveals that deep learning features effectively capture complex, high-level abstract textures that indicate the presence of pathologies. At the same time, radiomic features provide key information about the phenotypic characteristics of tumors, such as shape, texture, and intensity. This comprehensive approach not only improves the accuracy of non-invasive diagnostics, but also contributes significantly to personalized medicine by facilitating the development of more precise treatment strategies based on genetic profiles.