IMPROVED CONVOLUTIONAL NEURAL NETWORK BASED SEGMENTATION AND DETECTIONOF SKIN CANCER FROM DERMOSCOPY IMAGES USING MSER DESCRIPTOR

Segmentation of skin lesion from a dermoscopic images is a predominant footstep in computerized analysis approaches. Inaccurate skin lesion region segmentation could unfavorably impact the successive processing phases of anautomated skin cancer diagnosis system based on computer-aided because in these days, skin cancer is the most predominant forms of cancer diseases for descendant and light-skinned people.The most malignant type of skin cancer is “Basal Cell Carcinoma (BCC)”and in medical science, classification of BCC in earlier stage is a biggest issue for researchers. In the wake of patient, early detection of BCC is being a curable and useful of cancer diagnosis. On the way toachieve this significant goal, we designed a model for BCC classification that combines optimization segmentation approach with Convolutional Neural Network (CNN). Segmentation is a major concern of this research because skin lesion region extraction from dermoscopic image has a critical role in the early and accurate diagnosis of BCC cancer. However, automatic segmentation of skin lesions in dermoscopic images isa challenging task owing to difficulties including artifacts (hairs, gel bubbles, ruler markers), indistinct boundaries, low contrast and varying sizes and shapes of the lesion images. This paper proposes a novel and effective approach for skin lesion segmentation in dermoscopic images combining a deepconvolutional neural network with K-means and Cuckoo Search Algorithm (CSA). This research performs several steps for skin lesion segmentation using a dermoscopic image which is known as Region of Lesion (ROL) and used steps are: 1. Removalof hairs from images, 2. Exact lesion location detection, 3. K-means with CSA for segmentation of ROLas a foreground by subtracting the background data, 4. At last morphological operators are plied for post processing. The developed architectureis evaluating on publicly available and wellknown ISBI Datasets. When the evaluation parameters of proposed segmentation and detection of skin cancer work is compared with a few other existing state-of-art methods, the proposed method achieves the best performance of 98.1% in terms of Area Under the Curve (AUC) in differentiating BCC from benign lesions using only the Maximally Stable Extremal Regions (MSER) Features.