Second International Conference on Computational Intelligence in Data Science (ICCIDS-2019) 978-1-5386-9471-8/19/$31.00 ©2019 IEEE Performance analysis of Convolutional Neural Network (CNN) based Cancerous Skin Lesion Detection System G.S.Jayalakshmi, Department of Electronics Engineering , Madras Institute of Technology, Anna University, Chennai, India. jayalakshmigs.ece@gmail.com V.Sathiesh Kumar, Department of Electronics Engineering, Madras Institute of Technology, Anna University, Chennai, India. sathieshkumar@annauniv.edu Abstract— This paper focuses on the classification of dermoscopic images to identify the type of Skin lesion whether it is benign or malignant. Dermoscopic images provide deep insight for the analysis of any type of skin lesion. Initially, a custom Convolutional Neural Network (CNN) model is developed to classify the images for lesion identification. This model is trained across different train-test split and 30% split of train data is found to produce better accuracy. To further improve the classification accuracy a Batch Normalized Convolutional Neural Network (BN-CNN) is proposed. The proposed solution consists of 6 layers of convolutional blocks with batch normalization followed by a fully connected layer that performs binary classification. The custom CNN model is similar to the proposed model with the absence of Batch normalization and presence of Dropout at Fully connected layer. Experimental results for the proposed model provided better accuracy of 89.30%. Final work includes analysis of the proposed model to identify the best tuning parameters. Keywords—Dermoscopic Image Analysis, Deep Learning, Convolutional Neural Networks, Batch Normalization, Skin Lesion, Skin Cancer. I. INTRODUCTION Skin lesion refers to any abnormalities on the skin that may be benign or malignant. Malignant lesions are cancerous. Early diagnosis and treatment of precancerous skin lesion prevents the risk of cancer. WHO (World Health Organization) Statistics states that globally between 2 and 3 million non-melanoma skin cancers and 132,000 melanoma skin cancers occurs each year [1]. An additional 3,00,000 non-melanoma and 4,500 melanoma skin cancer case will result due to 10% decrease in ozone levels [1]. Generally, visual inspection of dermoscopic images requires a well-trained dermatology specialist. Due to Intraclass variation (colour, texture, shape and location) and Interclass Visual similarities of skin lesions, the diagnosis of skin lesion is a challenging task even for an expert Physician. Hence, Invasive biopsy of the affected lesion is required for accurate diagnosis by doctors. A Dermoscopic imaging technique utilizes non-polarized light source and a magnifying optics to capture and visualize deep skin structures. Unnecessary biopsies can also be avoided by analysing dermoscopic images prior to treatment as it provides a non-invasive method for skin cancer detection. With the emergence of deep learning approaches, intelligent medical imaging based diagnosis system can be developed. Deep learning technique allows the monitoring of the lesion automatically instead of frequent regular visit to dermatologist. Convolutional Neural Networks (CNNs) provide efficient classification mechanism across many fine- grained object categories with highly variable tasks [2]. This paper focuses on Classification of skin lesion types based on Batch normalized Convolutional neural networks. Advancement in computational hardware and availability of open source dermoscopic dataset makes Neural Networks as an efficient method for image classification. II. RELATED WORK Traditional lesion detection methods rely on hand- crafted features based on ABCDE rule (Asymmetry, Border, Color, Dermoscopic structure and Evolving) [3], 3-point checklist [4], 7-point checklist [5], Menzies method [6] and CASH (Color, Architecture, Symmetry, and Homogeneity) [7]. Numerous research works has been carried out to identify the skin lesion. These methods include hand-crafted feature extraction methods, conventional machine learning algorithms and deep learning techniques. • Lequan Yu et al. [8] proposed fully convolutional neural network for segmentation and deep residual network for classification. To cope with the degradation and over- fitting problems as the network goes deeper residual learning techniques were utilised [8]. Segmentation was required due to limited amount of dataset around 1200 images. Though 85.5% accuracy was achieved using residual technique and fully convolutional network, computational complexity is greatly increased. • Arkadiusz Kwasigroch et al. [9] focused their research on the application of transfer learning the existing efficient convolutional neural networks (CNN) which is a kind of Deep Neural Network (DNN), for automatic classification of the skin lesions. VGG19, Residual Networks (ResNet) and the hybrid of VGG19 with the Support Vector Machine (SVM) [9] are the three main CNN that were analysed. These CNN architectures were analysed and validated over the ISIC dataset. Modified VGG19 provided best performance with an average accuracy of 81.2% compared to VGG19-SVM and ResNet50 with 80.7% and 75.55% accuracies respectively.