Skin Melanoma Detection Using Image Augmentation

In order to effectively treat skin malignancy, a fatal kind of skin cancer, early detection and precise diagnosis are essential. Deep learning approaches have recently demonstrated promising outcomes in automating melanoma identification through examination of dermoscopic images. In contrast to comp...

Full description

Saved in:
Bibliographic Details
Published in2023 4th International Conference on Smart Electronics and Communication (ICOSEC) pp. 1624 - 1630
Main Authors Sivabalaselvamani, D., Nanthini, K., Selvakarthi, D., Niranchan, V.M., Kumar, L. Saravana, Swetha, P.
Format Conference Proceeding
LanguageEnglish
Published IEEE 20.09.2023
Subjects
Classification
Computational modeling
Computer architecture
Convolutional Neural Network
Deep learning
Dermascopic images
EfficientNet b3
Feature extraction
Melanoma
Predictive models
Skin melanoma
Training
Online AccessGet full text

Cover

More Information
Summary:In order to effectively treat skin malignancy, a fatal kind of skin cancer, early detection and precise diagnosis are essential. Deep learning approaches have recently demonstrated promising outcomes in automating melanoma identification through examination of dermoscopic images. In contrast to computer-aided diagnosis technologies, the manual diagnosis of melanoma images by dermatologists using dermoscopy is time-consuming and prone to error. Incorporating attention mechanisms in the CNN architecture can help the model focus on important regions within the dermoscopic images, potentially enhancing its ability to identify critical melanoma characteristics. Deep learning models like CNNs are often considered black-box models, making it difficult to interpret their decisions. Understanding how the model arrives at its predictions is crucial for gaining trust from medical professionals and ensuring patient safety. A convolutional neural network (CNN) architecture used in the suggested method was trained using both the original and enhanced dermoscopic pictures. Augmentation techniques such as rotation, scaling, flipping, and noise injection are applied to generate diverse samples, mimicking real-world variations in skin lesion appearance. By training the CNN model on this augmented dataset, it learns to extract meaningful features and generalize better, resulting in improved accuracy of 99.7% using EfficientNet b3.
DOI:10.1109/ICOSEC58147.2023.10275944