Skin Cancer Detection from Dermatoscopic Images through Integrated Deep Features and Ensemble Classification

The increase number of skin diseases year by year and their variety cause some difficulties in identifying the disease and classifying them. Skin diseases exhibit different sizes, shapes and visual characteristics that require an individual treatment strategy for each type of disease. Early detection of skin diseases is an important factor in preventing the adverse effects of the disease and treating them. Today, a lot of scientific researches are being conducted to solve this problem. However, the detection of skin diseases remains inefficient, imprecise and labor-intensive in manual procedures, emphasizing the need for automated methods. This research work presents an effective approach for early detection of skin diseases based on medical images and minimizing dependence on manual intervention. In the proposed method, the guided filtering techniques were combined with anisotropic Gaussian side windows (AGSW) to improve image clarity. On the basis of morphological analysis, regions of non-disease were removed from images, and with the help of deep neural networks, images were segmented, extracting high-quality regions of interest (ROIs) and multidimensional features. The proposed methodology used a soft voting classifier on the Random Forest (RF), Support Vector Machine (SVM), and K-Nearest Neighbor (KNN) models to categorize skin diseases identified from images. The proposed technique achieves an overall accuracy of 98.67% on the publicly available ISIC 2019 skin image dataset.

Ҳали таржима қилинмаган