An Efficient Skin Cancer Classification Approach Using Neural Networks

Abstract

Skin cancer is the most extensive cancer identified in the world and is highly contagious. Hence it is essential to prognosticate the disease at the earliest stage. The task is to forecast seven ailment categories with skin lesion images, containing actinic keratoses, melanocytic nevus, basal cell carcinoma, melanoma, and intraepithelial carcinomae, benign keratosis, dermatofibroma and pyogenic granulomas and hemorrhage. To accomplish the automatic categorization of skin lesions based on the disease, various neural networks such as Convolutional Neural Networks, Region-based Convolutional Neural Network and ResNet algorithms are used here owing to its fine-grained variability in the classification of skin lesions. The models are trained and assessed on dermatoscopic images collected from a cluster off publicly available datasets gathered manually. Initially exploratory data analysis is done over the images, followed by data pre-processing, augmentation, segmentation and extraction to contemplate the facts required to train the models and also to ensure the accuracy. This fully enhanced image is now fed into the algorithms where the features and bounded-regions are applied to get the convoluted layer. The model architecture generates a chain of layers to conclude on the output layer of seven neurons each indicating every class of disease. The presentation of the model is analysed based on various parameters such as precision, support, F1-score, accuracy etc. With these evaluation metrics a comparative study is done to judge the model behaviour. Also, the models are further assessed by making it to predict the class of disease of some random lesion. Hence the project is successfully created with help of requirement analysis, project plan, identifying features and functionalities, system validation and deployment. Thus, this application would act as the efficient need of the hour to obtain exact outcomes for the health care organization to act on instantaneously.