CNN Implementation on Major Skin Cancer Types Classification and NLP Diagnose Robot System

Abstract

Skin cancer, abnormal skin cell development, is a common and fatal type of cancer that occurs when skin is exposed to sunlight. Early diagnosis is important to prevent more serious consequences. Implementing a detection system would save more time for doctors and give patients efficient and low-cost diagnoses. In this paper, we built a skin cancer classification system based on Convoluted Neural Network (CNN) for seven majority skin cancers, and Natural Language Processing (NLP), for interaction with a human. We also implemented self-defined CNN, LeNet5, AlexNet, ResNet, VGG-16 in our system to compare their accuracy and discover reasons behind those output data. Finally, our self-defined CNN gets 0.8237 testing accuracy after training, LeNet5 results in 0.4857 testing accuracy, AlexNet produces 0.4715 testing accuracy, ResNet yields 0.8995 testing accuracy, and VGG-16 shown 0.7544 testing accuracy. The result indicates that ResNet-18 performs best through all models.