A Novel Approach for Detection of Skin Cancer using Back Propagation Neural Network

Author Name(s): Ms. Jignyasa Sanghavi
Author Email: sanghavijb1@rknec.edu

Abstract

Skin cancer is one of the most occurring cancers in the western countries. In US, millions of patients are reported for skin cancer. There are two main types of skin cancer Melanoma and Non melanoma skin cancer. The non melanoma skin cancer are reported occurring due to the exposure of skin to harmful ultra-violet (UV) rays, the rays internally disturbs the skin cell structure, causing malfunctioning of the skin cells. Early detection of skin cancer can reduce the mortality rate of skin cancer. Considering the seriousness of this domain, we tried to develop a system which will automatically detect the Melanoma skin cancer from skin lesion images. Advanced Imaging techniques are used for pre-processing, segmentation and feature extraction. Back propagation Neural network is used for classification. However now a days Deep neural networks are on boom, but they require huge data for training the networks. In medical domain such a huge data is not available for training. On comparison w i t h t h e results of the proposed system with some standard classifiers, we observe that the hybrid neural network classifier improves the accuracy by 15%.

Introduction

Skin cancer is curable if detected at early stage. Like many cancers, a skin cancer begins with normal visible lesions. These lesions are changes in skin that are not cancer, but can become cancer over time. According to the article of Forbes ,the nations which are influenced by ultraviolet radiations are United States, Canada, Cuba, parts of the Pacific and Northern Europe, South Asian nations, including India, Maldives, Bhutan and Bangladesh [1]. Skin cancer can be mainly categorized as three types such as Basal cell carcinoma (BCC), Melanoma, and Squamous cell carcinoma (SCC). The non-melanomas include BCC and SCC. BCCs are uncontrolled growth of lesions that rise in basal cells in the skin. Squamous-cell carcinoma (SCC) is a cancer which occurs in squamous cell. It spreads faster to other parts in the body. The lethal form of skin cancer is melanoma which arises in the melanocytes cell. The patients recovered from melanoma if it was diagnosed and treated in early stages. Many researchers tried to give solution for detection of Skin cancer using Image processing. The traditional method is consulting dermatologist. To develop the accurate system for skin cancer detection, sequencing of processing blocks is arranged according to methods of detection of dermatologist is following. Generally the steps followed are Preprocessing for noise removal, segmentation of region of interest, Feature extraction, training of classifier and finally classification. In the following section we tried to present the review of the researchers who worked on skin cancer detection using image processing, followed by the proposed method and it’s comparison with some standard classifiers. At last concluded with some interesting observations about the proposed classifier, and suggest ways on how it can be improved further.

Conclusion 

From the proposed work we conclude that the Backpropagation neural network based implementation is efficient method for skin cancer detection if more dataset is not available for training deep networks. It can perform better, and thus can be used as second opinion by health practioners for skin disease detection. As an extension to this work, the patients’ history can be inculcated with this system so as to give more accurate results.

732 total views, no views today

Download PDF File

About the author: admin