Detector quasi-periodic texture segmentation method for dermatological images processing

Abstract

Currently, digital diagnosis systems that process medical images are widely used in the diagnosis process in the field of healthcare. The purpose of such systems is to assist the doctor in establishing the diagnosis, or in monitoring changes in the patient's condition during treatment. Dermatology is one of the areas of medicine where the number of visits to a doctor is high. At the same time, the tasks of establishing a diagnosis and monitoring changes in the patient's condition during treatment are time-consuming and subjective and they depend on knowledge and experience of a dermatologist. However, today, digital systems for the diagnosis of dermatological diseases are not in every locality, expensive and are stationary systems. With the development of mobile information technologies, it became possible to develop mobile image processing systems for the analysis of dermatological diseases, which allow you to: receiving, analyze, and compare images before and after treatment at anytime, anywhere. One of the basic procedures in image processing systems is segmentation, the purpose of which is to reduce the amount of processed data. Segmentation methods can be classified as boundary-based methods and region-based methods. Dermatological disease images consist of regions which have difference by texture, that is, the segmentation problem is considered as the task of selection homogeneous regions by texture. The result of image processing depends on the quality of segmentation. To improve the quality of segmentation, in this work, we developed a detector quasi-periodic texture segmentation method for dermatological images processing, which contain quasi-periodic textures on a complex background in noisy conditions. This method is developed on the basis of the methodology of texture segmentation using detector, the stages of which are localization of spatial frequencies, detection, and contour segmentation. To localize of spatial frequencies, a wavelet-function improved by transform of graph of power function was used, which increases the accuracy of determining the boundaries of quasi-periodic textures contained in dermatological disease images. On the detection step, the comb filters, which are wavelets with a periodic or quasi-periodic transfer function that are applied to each image line, were used. The Canny method was use, as a contour preparation. Detector segmentation methods are focused on the image model. Therefore, a mathematical model of medical dermatological disease images was proposed, which contain quasi-periodic textures on a complex background in noisy conditions, as a model of a texture image with amplitude-modulated fluctuations in the intensity values by a random change in the amplitude and frequency of the oscillation. The developed method was applied to test medical images of psoriasis disease, which are available on the Internet. The accuracy of the segmentation of medical images of psoriasis disease containing a quasi-periodic texture was evaluated using the proposed method and the method using Gabor filters. It is shown that the proposed method is characterized by high speed and high segmentation quality, that is, it can be used in the development of express-diagnostic systems for monitoring changes in the patient's condition during treatment and to determine a parameter such as lesions area.