METHODS FOR DIAGNOSIS OF MELANOMA BASED ON DIGITAL IMAGE PROCESSING AND EXPERT SYSTEMS

Abstract

This article proposes a solution to a pressing scientific and applied problem: the development of a method for diagnosing melanoma based on a fuzzy expert system and digital image processing. The method for calculating melanoma features based on digital image processing proposed in the article includes: conversion of a color image into a grayscale image; conversion of a grayscale image into a binary image based on single-level global thresholding using the Otsu threshold; removal of small objects from the binary image using morphological transformation; formation of a binary matrix of image point membership to the object and a grayscale image of the object; determination of the object boundary in the binary image after morphological transformation based on the Kanna method; calculation of the irregularity of the object boundary in the binary image after morphological transformation; determination of the number of colors based on clustering of the gray-scale image object; rotation of the gray-scale image object; calculation of the diameter of the rotated gray-scale image object; verification of asymmetry based on the rotated gray-scale image object. For the diagnosis of melanoma, this work improved a fuzzy expert system for melanoma diagnosis that uses Sugeno’s fuzzy inference algorithm. An experimental study confirmed that the proposed fuzzy expert system achieves a probability of incorrect decisions regarding melanoma diagnosis of 0.02 and a root mean square error of 0.05. The scientific novelty of the study lies in the fact that the proposed fuzzy expert system represents knowledge about melanoma in the form of fuzzy rules that are understandable to humans; it reduces computational complexity, the probability of making an incorrect decision, and the root mean square error. The proposed solution is scalable and suitable for use in intelligent decision-making systems.