Melanoma Classification Using Machine Learning and Deep Learning

This paper investigates the application of machine learning algorithms, including ResNet18, for enhancing melanoma detection using the ISIC2020 dataset, comprising two classes: benign and malignant. In addition to traditional classification algorithms (SVM, random forest, KNN, and logistic regressio...

Full description

Saved in:

Bibliographic Details
Published in:	2023 1st International Conference on Health Science and Technology (ICHST) pp. 1 - 6
Main Authors:	Vu, Tran Anh, Son, Pham Quang, Hiep, Dinh Nghia, Huy, Hoang Quang, Kien, Nguyen Phan, Viet Huong, Pham Thi
Format:	Conference Proceeding
Language:	English
Published:	IEEE 28-12-2023
Subjects:	Accuracy Deep learning Feature extraction Histogram of Oriented Gradients Logistic regression Machine learning Machine learning algorithms Melanoma Support vector machines
Online Access:	Get full text
Tags:	Add Tag No Tags, Be the first to tag this record!

Description
Summary:	This paper investigates the application of machine learning algorithms, including ResNet18, for enhancing melanoma detection using the ISIC2020 dataset, comprising two classes: benign and malignant. In addition to traditional classification algorithms (SVM, random forest, KNN, and logistic regression), ResNet18, a deep learning convolutional neural network, was employed, showcasing an impressive accuracy of 96.98%. The study focuses on feature selection using the Histogram of Oriented Gradients (HOG) method and applies preprocessing techniques to optimize data quality. HOG features are extracted to represent skin lesions, and all algorithms are trained and evaluated using cross-validation, yielding accuracy rates ranging from 84 % to 96.98%. The study highlights the potential of machine learning, including HOG-based feature selection and ResNet18, for accurate melanoma detection. Integrating these models into clinical practice could significantly improve diagnostic accuracy and aid in treatment decisions. Further research directions involve exploring ensemble techniques, addressing interpretability concerns, and investigating the synergistic potential of combining traditional algorithms with deep learning models. Overall, this study demonstrates the effectiveness of HOG-based feature selection and ResNet18 in enhancing melanoma detection, underscoring the promise of machine learning in advancing patient outcomes.
DOI:	10.1109/ICHST59286.2023.10565364