Machine Learning of Skin Cancer Diagnosis Based on Oriented FAST and Rotated BRIEFS

Abstract

Screening for malignant melanoma is critical to treating the disease and saving lives. Many computerized techniques have been reported in the literature to diagnose diseases and classify them with pathological performance to detect skin cancer. However, reducing the rate of misrepresentation remains a challenge and a concern, the use of false positives is worrying and requires intervention by an expert pathologist for further examination and examination. In this paper, an automated skin cancer diagnosis system that combines different properties that use new features in texture and color has been proposed in a distinctive bag approach for effective and accurate detection. To give good results, images must be processed by using an average filter in order to eliminate the noise that may occur in the image, so we need to highlight the affected part, so it is isolated through the watershed method and then the affected part is distinguished using the colored skin lesion hash algorithm that teaches representative texture distributions and calculates the texture excellence scale for each distribution. A fabric vector is extracted for each pixel in the image. Powerful features are then extracted using the ORB method.  Finally, to give good results, one method of machine learning is SVM, which is one of the distinct ways of efficiency, speed and accuracy compared to other methods.      The system was trained and tested at 30% of the HAM10,000 data set used, so when calculating the system's assessment of skin cancer, it was discovered that the results were excellent, with 98% accuracy, 95% sensitivity, and 96% privacy, meaning that the system works accurately and efficiently.