Classification of Skin Lesion by Interference of Segmentation and Convolotion Neural Network

Classification of skin lesions plays a crucial role in diagnosing various, local and gene related, medical conditions in the field of dermoscopy. Estimation of these biomarkers are used to provide some insight, while detecting cancerous cells and classifying the lesion as either benign or malignant. This paper presents groundwork for detection of skin lesions with cancerous inclination by segmentation and subsequent application of Convolution Neural Network on dermoscopy images. Images included in ISIC-2016 were used as dataset. Images with skin lesions were segmented based on individual channel intensity thresholding. The resultant images were fed into CNN for feature extraction. The extracted features were then used for classification by an ANN classifier. Previously, several approaches have been used for subject diagnostic with varying degree of success. However, room is still available for exploring other techniques for improving proportion of successfully detected malignant lesions. As compared to a previous best of 97%, methodology presented in this paper yielded an accuracy of 98.32%.