Comparative study on different Deep Learning models for Skin Lesion Classification using transfer learning approach

Code

• github.com/Abhishek-st/Skin-Lesion-Analysis
  In papertf★ 2

Abstract

Developing countries, specifically India, do not have sufficient hospitals and doctors to reach out to the population. Forget about skin specialists, there are still thousands of villages without even a basic hospital. But, there is one thing that reaches out to every person in this country which is the internet. This research paper focuses on training different pre-trained models on our dataset and suggests the best one for skin lesion classification. This model has been used with a mobile compatible web app to detect skin cancer at the earliest stage possible. Hence, the objective of this research paper is to develop an intelligent system to detect skin cancer at the earliest stage possible by a skin lesion classifier. The dataset includes different categories of diseases like Actinic Keratosis, Vascular Lesions, Lentigo, Melanoma, and Dermatofibroma. The pre-trained models used in this research are the different transfer learning methods like VGG19, Xception, Densenet, Inception, MobileNet, NasNetMobile, and Resnet. Using transfer learning, Deep Neural Networks can be trained with presumably less amount of data. Also, transfer learning has been consistently proven to reduce training time as well as boost model accuracy. Hence, this paper has given the emphasis on using transfer learning models instead of building a model from scratch.

Tasks

Reproductions