Ana Kurniawati, Aniqoh Hana Haura

Comparative Analysis Of Convolutional Neural Network Models For Digital Image-Based Melanoma Classification

Introduction

Comparative analysis of convolutional neural network models for digital image-based melanoma classification. Compare CNN models for digital image-based melanoma classification. Address invasive diagnosis with a rapid, non-invasive solution. Achieved 93.18% accuracy, supporting early detection.

Abstract

Melanoma is one of the most malignant forms of skin cancer, with an incidence rate of 7.9% in Indonesia. Traditional biopsy-based diagnosis, though crucial, is invasive and time-consuming, creating barriers for early detection. To address this issue, this research compares two Convolutional Neural Network (CNN) models for digital image-based melanoma classification. The study utilized a publicly available dataset from Kaggle, consisting of 17,805 images (melanoma and non-melanoma), which were divided into training, validation, and testing subsets. The models were trained using the Adamax and SGD optimizers for 100 epochs. The performance of the models was evaluated based on accuracy, loss, precision, recall, and F1-score. The CNN model with the best architecture, which consisted of two fully connected layers, achieved an accuracy of 93.18% and a loss of 0.1636, outperforming the alternative model. These results confirm the effectiveness of CNN models in classifying melanoma images and support the development of a web-based platform that allows users to upload or capture images for rapid and non-invasive detection.

Review

This study addresses a critical challenge in public health: the early and non-invasive detection of melanoma, a highly aggressive form of skin cancer. Given the high incidence rates, such as 7.9% in Indonesia, and the limitations of traditional biopsy-based diagnosis—which is invasive and time-consuming—the need for efficient screening tools is paramount. The authors propose a timely and relevant approach by exploring the utility of Convolutional Neural Networks (CNNs) for digital image-based classification of melanoma, aiming to overcome existing barriers to early detection. The research meticulously compares two distinct CNN models, leveraging a substantial publicly available dataset from Kaggle comprising 17,805 images of both melanoma and non-melanoma cases. This dataset was appropriately partitioned for training, validation, and testing, a standard practice for robust model evaluation. Training involved common optimizers, Adamax and SGD, over 100 epochs, and model performance was comprehensively assessed using key metrics including accuracy, loss, precision, recall, and F1-score. A notable finding is that the CNN model employing a specific architecture, characterized by two fully connected layers, achieved a commendable accuracy of 93.18% and a low loss of 0.1636, significantly outperforming the alternative model under comparison. The strong performance demonstrated by the superior CNN model underscores the significant potential of deep learning methodologies in enhancing dermatological diagnostics. These results not only confirm the effectiveness of CNNs for classifying melanoma images but also lay a solid foundation for practical application. The authors rightly highlight the prospect of developing a web-based platform, which would empower users with a rapid, non-invasive tool for initial melanoma screening. While the abstract provides strong evidence of technical proficiency and promising outcomes, a full manuscript would benefit from a deeper dive into the specific architectures of the compared CNN models and a discussion of the factors contributing to the superior performance of the chosen architecture. Nevertheless, this work represents a valuable contribution towards revolutionizing early melanoma detection.

Full Text

Comments

You need to be logged in to post a comment.