A Comparative Analysis of Transfer Learning-based Techniques for the Classification of Melanocytic Nevi

TL;DR

This paper compares five transfer learning techniques based on deep convolutional neural networks for classifying melanocytic nevi, aiming to improve early skin cancer detection.

Contribution

It provides a comparative analysis of pre-trained DCNN-based transfer learning methods for skin lesion classification, highlighting their potential for medical diagnosis.

Findings

Identifies the most effective transfer learning model for skin lesion classification.

Demonstrates the potential of deep CNNs in early skin cancer detection.

Provides insights into the strengths and limitations of each technique.

Abstract

Skin cancer is a fatal manifestation of cancer. Unrepaired deoxyribo-nucleic acid (DNA) in skin cells, causes genetic defects in the skin and leads to skin cancer. To deal with lethal mortality rates coupled with skyrocketing costs of medical treatment, early diagnosis is mandatory. To tackle these challenges, researchers have developed a variety of rapid detection tools for skin cancer. Lesion-specific criteria are utilized to distinguish benign skin cancer from malignant melanoma. In this study, a comparative analysis has been performed on five Transfer Learning-based techniques that have the potential to be leveraged for the classification of melanocytic nevi. These techniques are based on deep convolutional neural networks (DCNNs) that have been pre-trained on thousands of open-source images and are used for day-to-day classification tasks in many instances.