Artificial Intelligence Help Classify Eye Cancer

Usually, eye cancer may affect both the internal and external parts of the eye, for instance, the eyelid and the conjunctiva. According to Ziaei et al. (2006), eye cancer is an abnormal cell growth or tumor in any part of the eye which could be primary or secondary. In the case of primary tumors, the abnormal growth begins from the eye parts affected while in secondary cases, the cancerous cells result from ailments in other parts of the body before spreading to the eye (Siegel et al., 2006).

eye cancer may begin inside the eyeball. In such cases, it is referred to as intraocular malignancy, argues researchers. The commonest intraocular tumor is uveal melanoma which occurs in adults. According to Siegel et al. (2006), the tumors may occur in the iris, ciliary and choroid. the eye cancer has a list of risk factors, chief among them, fair complexion which includes a fair skin that can easily burn. Other factors include white skin and old age. According to a guidance by Rundle (2017), The symptoms of eye cancer include blurred vision, flashes or floaters of light, a dark spot in the iris, change in size or shape of the pupil and alterations of the eyeball’s position.

Recently, Deep Learning (DL) starts to play an essential role in data science. It has a layered-based architecture that has been motivated by artificial intelligence. These layers apply a nonlinear transformation that helps to extract the best features for classification problems. There have been different kinds of classification problems in DL methods which showed state-of-the-art results on different image recognition datasets in terms of the reduction of the error rate. 

many studies have focused on the classification of Deep Learning techniques over-supervised and unsupervised data sets. The goal of all these studies is to get the better classification accuracy and less error rate. This paper will focus on classification problems of deep learning techniques over supervised data sets. Following a list of some of the related works:

Li, Zeiler, Zhang, Yann, and Rob have introduced a new algorithm in the Deep Learning field called DropConnect which is a generalization of Dropout. Their algorithm has a similar concept of the Dropout algorithm. However, the DropConnect algorithm has a new technique which is randomly dropping out some of the weights instead of randomly dropping out some of the activations as it's in Dropout algorithm to avoid the over-fitting problem. This technique shows a significantly better result in classification problems with fully-connected layers in Neural Network. They have used both of the algorithms in several datasets such as MNIST, CIFAR-10, SVHN, and NORB to check the accuracy and error rate. DropConnect algorithm showed state-of-art results on these datasets by getting less error rate among all the previous studies.

This paper is structured as follows: the first section is an introduction to the eye tumors and the proposed system. The second one is a methodology of the system in which we provide flowcharts and figures that explain the system stages and the

methods used. In the subsections of the second part, we explain each image processing technique that is used in the system. Section three is the experimental results which show the system efficiency. The last section is a conclusion about the developed approach.

Cite this page