|
Skin lesion classifier: A modern approach using convolutional neural networks
|
|
| Author:
|
MOVVA JAYA SURYA, MOVVA LALITH , A. KRISHNA MOORTHY, V. VIJAYARAJAN , R. KANNADASAN , P. BOOMINATHAN
|
| Abstract:
|
Among the prevailing diseases on earth, cancer is one of the most dangerous one. To cure the cancer become
difficult if the cancer cells persist in the body for a long time, so it is better to identify the cancer cells early
and to start the treatment but some backward places do not have the accessibility for proper technology to
identify the symptoms of cancer cells, especially for skin cancer which is increasing rapidly in the recent years
due to increase in UV radiations that are reaching earth directly. Since experienced dermatologists are not
mostly available in backward regions, Implementation of artificial intelligence can take place for detecting skin
cancer in such remote places with non-professional dermatologists. Since the deep learning models are
achieving human level accuracy in visualising the tasks and sometimes even with more accuracy than humans in
the cases where things cannot be visualised by naked eye and where it is difficult for humans to classify. In this
paper, A deep learning model is constructed which classifies the given image into any of the following seven
categories. They are melanocytic nevus, dermatofibroma, melanoma, basal cell carcinoma, actinic keratosis,
vascular lesion and benign keratosis. This model is trained on the HAM10000 dataset which is provided by an
organization namely “The International Skin Imaging Collaboration (ISIC)”. This model can be trained and also
be loaded into a mobile application which may have high resolution camera connected to it. So it can take the
image of the patient skin where it is infected and feed the image as the input to our proposed deep learning
model for evaluating the image as well as classifies the image to its particular class with high accuracy.
|
| Keyword:
|
Convolutional Neural Networks, DenseNet201, Dermoscopy, Tensor flow.
|
| EOI:
|
-
|
| DOI:
|
-
|
| Download:
|
Request For Article
|
|
|