A Convolutional Neural Network-Based Web Prototype to Support Melanoma Skin Cancer Detection

The skin is the organ that protects the human body. However, factors such as solar radiation damage the texture and skin cells. Sometimes, the lack of timely diagnosis leads to skin cancer. In this line, melanoma is the most dangerous type of cancer and has caused the greatest number of deaths related to skin diseases. With this problem, collaborative efforts between different research areas are necessary to support early detection of the disease. In this way, the evolution of algorithms based on neural networks plays an important role for image processing, which is an essential activity for pattern detection and recognition in medical diagnosis. Faced with this challenge, this research proposes a web prototype based on convolutional neural networks to support melanoma detection. In this context, the reference framework suggested by the Cross Industry Standard Process for Data Mining (CRISP-DM) was used. For this, 18,000 high-quality images were compiled from the data science community. In addition, two learning models (based on convolutional neural network and Res-Net50) were created and evaluated. With these premises, a web application was developed using the waterfall model. Finally, conclusions and future work are suggested at the end of the document.