Skin Cancer Classification Using Convolution Neural Networks

Abstract

The incidence of skin cancers has been increasing over the past decades at an alarming rate. Right now, somewhere in the range of 2 and 3 million non-melanoma skin diseases and 132,000 melanoma skin malignant growths happen all-inclusive every year. One in each three diseases analyzed is skin malignant growth. As ozone levels keep being drained, the air loses increasingly more of its defensive channel capacity, and progressively sun-oriented UV radiation arrives at the Earth’s surface. It is assessed that a 10 percent decline in ozone levels will bring about an extra 300,000 skin malignant growth cases. Hence, skin cancer today poses a serious threat to mankind. One of the major reasons for skin cancer fatalities is the absence of an early diagnosis. When detected early, skin cancer survival exceeds 95%. Therefore, to facilitate the early diagnosis of skin cancer, we propose this solution—a machine learning model trained on HAM10000 dataset (a large collection of multi-source dermatoscopic images of common pigmented skin lesions) using convolutional neural networks (CNN), to classify a given skin lesion image into various cancerous (or non-cancerous) skin conditions. This model incorporated into an online platform will enable doctors and laboratory technologists to know the three highest probability diagnoses for a given skin lesion. Hence, the model will be of immense help in quickly identifying high priority patients and speeding up the procedural workflow.