A Hybrid Deep Learning Framework for Skin Cancer Classification using Dermoscopy Images and Metadata

Abstract

Skin exposed to the sun is more likely to have skin cancer, which is an unnatural growth of skin cells. However, skin that isn’t frequently exposed to sunlight might potentially develop this prevalent type of cancer. This study attempts to detect and categorize six different types of skin cancer using clinical images accurately. The proposed approach used both clinical images and Metadata to feed into a hybrid model that benefits from deep learning in feature extraction and machine learning in classification. Many experiments have been applied to the PAD-UFES-20 dataset. Seven distinct algorithms of machine learning are applied with ten different pre-trained deep learning models. The hybrid model consists of MobileNetV2 as a feature extractor and logistic regression as a classifier for images, and random forest as a classi-fier for metadata has proved its superiority. The accuracy was 95.6%, precision 96.8%, recall 95.6%, F1-Score 95.7%. The evaluation process shows that metadata has an effective role in improving accuracy compared to previous studies that depend on images only in classification.

SkinLesNet: Classification of Skin Lesions and Detection of Melanoma Cancer Using a Novel Multi-Layer Deep Convolutional Neural Network

“…In [ 52 ], the PH2 dataset of dermoscopic images was utilized to create and build a CNN model, which achieved test-set accuracy of over 95%. In [ 53 ], a six-layer CNN model was created and trained on the ISIC dataset and showed promise, with accuracy of 89.30% in classifying skin lesions. Another State-of-the-Art CNN model was designed and developed by [ 54 ].…”