Classification of skin cancer using VGGNet model structures

Abstract

Skin cancer is one of the most common type of cancer in humans. This type of cancer is produced by skin cells called melanocytes and occurs as a result of division and multiplication of the mentioned cells. The most important symptom of skin cancer is the formation of spots on the skin or the observation of changes in the shape, color, or size of the existing spot. It is necessary to consult a specialist to distinguish the difference between a normal spot and skin cancer. Expert physicians examine and follow up the spots on the skin using skin surface microscopy, called dermatoscopy, or take a sample from the suspicious area and request it to be examined in laboratory environment. This situation increases the cost of the procedure for the diagnosis of skin cancer and also causes it to be treated at a later stage. Therefore, there is a need for a metod that can detect skin cancer early. Thanks to machine learning, become popular in recent years, many diseases can be diagnosed with software that helps expert physicians. In this study, VGGNet model structures (VGG-11, VGG-13, VGG-16, VGG-19) that quickly classify skin cancer and become a traditional convolutional neural network architecture using deep learning method, a subfield of machine learning, were used. It has been observed that the VGG-11 architecture, which is one of the VGGNet model structures, detects skin cancer with superior success accuracy (83%) compared to other model structures.

Upper and lower extremity bone segmentation with Mask R-CNN

“…To assist doctors in detecting bone-related disorders, computer-aided diagnosis (CAD) has been widely used to analyze medical images and has received increasing attention [3]. Deep learning applications are intensively used to classify health data and can potentially provide pioneering knowledge to domain experts [4,5]. In our study, a segmentation application that provides the boundary information of the object with formal precision will be applied beyond visual classification and object identification within frames.…”