Efficient and Low-Cost Skin Cancer Detection System Implementation with a Comparative Study Between Traditional and CNN-Based Models

Lakindu Induwara Mampitiya; Namal Rathnayake; Subashini De Silva

doi:10.47852/bonviewJCCE2202482

Authors

Lakindu Induwara Mampitiya Department of Electrical and Electronic Engineering, Sri Lanka Institute of Information Technology, Sri Lanka https://orcid.org/0000-0002-4397-2526
Namal Rathnayake School of Systems Engineering, Kochi University of Technology, Japan https://orcid.org/0000-0002-5235-8552
Subashini De Silva Department of Electrical and Electronic Engineering, Sri Lanka Institute of Information Technology, Sri Lanka https://orcid.org/0000-0003-1700-0093

DOI:

https://doi.org/10.47852/bonviewJCCE2202482

Keywords:

HOG, MobileNet, PCA, ResNet50, skin cancer, SMOTE, SVM

Abstract

Medical image classification is an essential task in the field of combining medical applications with Artificial Intelligence. This study is carried out to introduce an accurate, precise method for skin cancer recognition. This research investigates the performance of classifying skin cancer dataset HAM10000 using ResNet50, MobileNet, and the traditional Support Vector Machines (SVM) model. The dataset combines seven cancer types: Actinic Keratosis, Basal Cell Carcinoma, Benign Keratosis, Dermatofibroma, Melanoma, Melanocytic Nevus, and Vascular Lesion. The SVM classifier is designed to employ a Histogram of Oriented Gradient (HOG) features with Principle Component Analysis (PCA). Moreover, the Synthetic Minority Oversampling Technique is used to balance the dataset. Additionally, six conventional machine learning (ML) methods are used to compare the results with the calculation of precision, recall, F1 Score, and accuracy. The results confirm that the SVM method outperforms the other algorithms with an accuracy of 99.15%. The novelty contribution of this research activity is mainly based on the development of a high accuracy, low computational complex machine method for skin cancer types recognition in the domain of medical image classification.

Received: 24 October 2022 | Revised: 5 December 2022 | Accepted: 12 December 2022

Conflicts of Interest

The authors declare that they have no conflicts of interest to this work.