Skin Cancer Classification Systems Using Convolutional Neural Network with Alexnet Architecture

Abstract

Skin cancer is one of three malignant cancers that are often found in Indonesia and can cause death, apart from cervical cancer and breast cancer. Skin cancer diagnosis done manually by a dermatologist through a biopsy and microscopic process. However, this process takes a long time and has a risk of accidents in the biopsy process. Whereas early diagnosis shows that more than 90% can be cured, while delay in diagnosis shows less than 50% can be cured. For this reason, this research was conducted to design and implement a system that can classify skin cancer in order to facilitate the skin detection process to be faster. This study applied the Convolutional Neural Network (CNN) method using the Alexnet architecture to classify skin cancers into four categories, namely dermatofibromas, melanoma, nevus pigmentosus, and squamous cell carcinoma. Experiments were carried out using a dataset obtained from the International Skin Imaging Collaboration (ISIC) dataset of 4000 images of skin cancer conditions of dermatofibroma, melanoma, nevus pigmentosus, and squamous cell carcinoma, consisting of 1000 images in each class. The amount of training data used is 2700 skin cancer images. While the number of validation data used is 900 images. Then the remaining 400 images were used as test data. The experimental results using training data and validation data show that the best optimizer is Adam and the use of a learning rate of 0.001 produces an accuracy value of 99% and a loss value of 0.0596 on the applied Alexnet architecture. From the Alexnet model, the test data were successfully detected according to their respective classes with the test data accuracy value of 97.5%, loss value of 0.1272 and the precision, recall and f1-score value around 0.96–0.99. Based on the results of the system performance, it shows that the applied model promises to be an early detection tool for skin cancer by dermatologists so that it can reduce the dangerous effect due to the delay in diagnosing skin cancer.