Texture segmentation method for computer-assisted dermatologic diagnostic system

Abstract

The wide spread of dermatological diseases is an important medical and social problem. Doctors note the constant growth of
psoriasis among people of all ages. The psoriasis disease symptoms are similar to the symptoms of such diseases as eczema, atopic
dermatitis and medicinal disease. Therefore, there is a high probability of an error in the disease diagnosis, which prevents the full
treatment and prevention of the disease. Dermatological diagnostic systems are decision-making support systems for dermatologists
when establishing a diagnosis and assessing the severity of the disease course. The development of new image processing methods
for dermatological diagnostic systems is an important task, which allow to increase the accuracy of the diagnostic decision. In this
work, the segmentation method of psoriasis images for systems of medical dermatological diagnostics based on a vector-difference
approach to improve the quality of segmentation was developed. The vector-difference approach allows to calculate a certain texture
feature of the image as a vector transformation of various texture features by linear algebra methods. Psoriasis disease images can be
described by texture (spectral, statistical, spectral-statistical) and color, so it is proposed to take into account textural and color
characteristics of images for image segmentation. The color models that are most often used in segmentation methods of psoriasis
disease images were analyzed. Based on the analysis, the Hue-Saturation-Intensity color model was chosen. It is proposed to use
spectral, statistical and spectral-statistical texture models and color characteristics of image pixels to represent psoriasis disease
images. The developed segmentation method includes the following stages: image pre-processing; identification; vector-difference
transformation; threshold processing. At the pre-processing stage, homomorphic filtering was applied to psoriasis disease images.
The result of the identification stage is a set of features calculated by the textural and color characteristics for image objects. The
vector-difference transformation converts the texture image into intensity. Threshold processing is performed with a global threshold.
Experimental research of the proposed segmentation method of psoriasis disease images was performed. As a result of the
experimental research, it was found that the probability of correct identification of psoriasis disease area on average is 0.97, the
probability of a false alarm is about 0.05.