An Efficient Skin Mole Segmentation Using Deep Convolutional Neural Networks (DCNNs)

Abstract

Every year, specialists analyze skin malignant growth in around 3 million Americans or
more. Presently it is one of the most widely recognized kinds of cancer. On the off chance
that skin malignancy is diagnosed early, it can without much of a stretch be treated with
topical drugs. So as a result, skin cancer is responsible for less than 1% of all cancer deaths.
There are two moles are found in skin disease observation, and those are benign and
malignant. So the early diagnosis of skin disease is important to forestall any serious
condition for the patient. Its conclusion is urgent if not distinguished in the beginning period.
The paper intends to recognize considerate and dangerous types of skin malignant growth
utilizing dermoscopic images, applying various deep convolutional neural network
algorithms. We utilize three different pre-trained deep learning models, in particular as
VGG19, ResNet50 and EfficientNetB0 to empower the most effective model to recognize skin
malignant growth conditions. Where VGG19 has 84.44%, ResNet50 has 94.29% and
EfficientNetB0 has 98.67% accuracy on the ISIC archive skin cancer dataset. We compared
ReLu and Softmax activation functions output in this experiment as well.

Keywords: skin cancer, deep convolutional neural network, benign, malignant, deep learning
techniques, segmentation