Disease Classification based on Dermoscopic Skin Images Using Convolutional Neural Network in Teledermatology System

• Main Authors: I Ketut Eddy Purnama, Arta Kusuma Hernanda, Anak Agung Putri Ratna, Inggrid Nurtanio, Afif Nurul Hidayati, Mauridhi, Hery Purnomo, Supeno Mardi Susiki Nugroho, Reza Fuad Rachmadi
• Format: Book Section PeerReviewed
• Language: English; English; English
• Published: IEEE Xplore Digital Library 2019
• Subjects: R Medicine (General); RL Dermatology
• Online Access: http://repository.unair.ac.id/95072/1/Disease%20Classification%20based%20on%20Dermoscopic%20Skin.pdf; http://repository.unair.ac.id/95072/2/Disease%20Classification%20based%20on%20Dermoscopic%20Skin%20Images.pdf; http://repository.unair.ac.id/95072/3/Disease%20Classification%20based%20on%20Dermoscopic%20Skin%20Images%20Using%20Convolutional%20Neural%20Network%20in%20Teledermatology%20System.pdf; http://repository.unair.ac.id/95072/; https://ieeexplore.ieee.org/document/8973303
• Institution: Universitas Airlangga

We have proposed a system of classification and detection of skin diseases that can be applied to Teledermatology. This system will classify skin diseases on dermoscopic images using the Deep Learning algorithm, Convolutional Neural Network (CNN). Dermoscopic image data in this study from MNIST HAM10000 dataset which amounts to 10,015 images and published by International Skin Image Collaboration (ISIC). The dataset is divided into seven class of skin diseases which fall into the category of skin cancer. The image classification process will use two pre-trained CNN models, MobileNet v1 and Inception V3. The model results from the learning process will be applied to a web-classifier. The comparison of predictive accuracy shows that the web-classifier using the CNN Inception V3 model has an accuracy value of 72% while the web-classifier that uses the MobileNet v1 model has an accuracy value of 58%.