Segmentation and Border Detection of Melanoma Lesions Using Convolutional Neural Network and SVM

Melanoma is one of the most lethal forms of skin cancer caused when skin is exposed to intense UV rays. Estimates suggest that the deaths tolls are more than 50,000 with 3 million and more reports of it yearly. However, early diagnosis of malignant melanoma significantly curbs the mortality rate. Several computer-aided diagnosis systems have been proposed in assisting the detection of malignant melanoma in its earlier stages. These systems help in early detection and earlier diagnosis of many symptoms, which results in better and accurate treatment. However, the challenge starts from the first step of implementation of such systems, which is melanoma lesion detection in the image. In this paper, the problem of automatic detection of melanoma lesion on skin images is presented based on the concept of deep learning. The experiments have been performed using Convolutional Neural Networks (CNNs) with training input size of 15 × 15 and 50 × 50. The result of the study shows that deep learning using CNN is able to detect the melanoma lesion efficiently. The best performance has been achieved using CNN with 15 × 15 training input size. The performances obtained using this network is Jaccard index (0.90), Accuracy (95.85%), Precision (94.31%), Recall (94.31%), and F-value (94.14%) for the best performance.