Melanoma Classification using Deep Learning Architectures and Transfer Learning

Abnormal growth of skin cells, especially Melanoma, one of the most serious types of skin cancer, is caused by melanin-producing cells (melanocytes). Melanoma can also form in the eyes and, rarely, be inside the body, such as in the patient's nose or throat. Melanoma is one of the most deadly diseases that can be successfully treated if it is diagnosed early. Many existing technologies have shown that computer vision can play a major role in the study of medical imaging. In this paper, we are identifying melanoma in lesion images using Ensemble learning under Deep Learning models. The proposed model forecasts the likelihood (floating point) that the lesion in the image is malignant between 0.0 and 1.0 with an accuracy of 98.1%. The number 0 signifies benign and 1 indicates malignant in the training data.

I. Introduction

Not-withstanding being the most not unusual skin cancer, cancer debts for seventy-five percent of all skin cancer deaths. In line with the American Cancer Society, approximately 100,000 new instances of melanoma could be diagnosed by 2022. Nearly 7,000 individuals are anticipated to die as a result of the sickness [1]. Early and accurate detection, as with other cancers, can improve treatment effectiveness, which could be improved by data science [11]. Dermatologists currently take a look at all of a patient's moles to discover outlier lesions or “ugly ducklings” that are maximum in all likelihood to be melanoma.