Biomedical Data Analysis Systems for the Diagnosis of Skin Neoplasms

Introduction. The effectiveness of the diagnosis of malignant neoplasms of the skin remains unsatisfactory due to the complex process of interpretation of clinical features. On the other hand, in the last two decades, noninvasive optical diagnostic methods have been actively developed, for example, digital dermatoscopy for visualization of surface neoplasms and Optical Coherence Tomography (OCT) for obtaining spatial scans. Recent advances in the study of non-invasive diagnostic tools makes this area very promising for research in a clinical condition. Aim. Developing of software modules based on the mathematical framework of texture analysis for biomedical data systems designed for the diagnosis of skin malignant neoplasms. Materials and methods. Algorithms of software modules developed for optical systems of our own design are presented. Algorithms for a dermatoscopic module are based on the Haar transform, Local Binary Patterns and color features. Algorithms for OCT are based on the texture features of Haralick, Tamura, fractal dimension, complex directional field and Markov random field. Studies were conducted on sets of 106 dermatoscopic and 1008 OCT images of various classes of pathologies, including melanoma and Basal Cell Carcinoma (BCC). Results. The values of sensitivity and specificity for the dermatoscopic system and OCT were experimentally obtained. Conclusion. The sensitivity of the dermatoscopic system is 90 % versus 93 % for other authors, as well as the specificity is 86 % versus 80 %. One of the factors of the increase can be considered the introduction of a personalized mode - the addition of comparative features evaluating a difference between a tumor and a normal tissue in the software analysis module. The improved accuracy of OCT is up to 97 % for the diagnosis of melanoma and up to 96 % for the diagnosis of BCC