Abstract
— Skin-cancer is the most common type of cancer among all
other types of cancers spreading both developed and developing
countries. In this paper, a web service is developed in order to help
physicians as well as individuals to upload an image and diagnose the
particular types of lesion. Computer vision is used to analyse lesions
on images by providing computers with somewhat similarity as humans do.
For this a Convolution Neural Network (CNN) is used with multi
classification on International Skin Imaging Collaboration (ISIC) 2018
dataset with HAM10000 images. This dataset is a meta-data which has
various types of images with seven different labels. At first, the model
is trained with a larger training set and saved in a zip folder.
Secondly, a web service is created where users or a doctor can upload an
image for classification. Thirdly, the images uploaded are pre-processed
as there is noise, hairs on image. Techniques like resizing,
normalisation, thresholding, black-hat filtering and inpainting are used
for this purpose. After this, the saved model is called to define
whether the uploaded image is benign or malignant. The experimental
results reveal that the proposed model is superior in terms of detection
and diagnosis accuracy as compared to modern methods.