Abstract:Background: Melanoma is a skin cancer that starts when the melanocytes that produce the skin color pigment start to grow out of control and form a cancer. Detecting melanoma early before it spreads to the lymph nodes and other parts of the body is very important because it makes a big difference to the patient's 5-year life expectancy. Screening is the process of conducting a skin examination to suspect a mole is melanoma using dermoscopic or macroscopic images. However, manual screening takes a long time. Therefore, automatic melanoma detection is needed to speed up the melanoma detection process. The previous studies still have weakness because it has low precision or recall, which means the model cannot predict melanoma accurately. The distribution of melanoma and moles datasets is imbalanced where the number of melanomas is less than moles. In addition, in previous study, comparisons of several CNN transfer learning architectures have not been carried out on dermoscopic and macroscopic images. Objective: This study aims to detect melanoma using the Convolutional Neural Network (CNN) with transfer learning on dermoscopic and macroscopic melanoma images. CNN with Transfer learning is a popular method for classifying digital images with high accuracy. Methods: This study compares four CNN with transfer learning architectures, namely MobileNet, Xception, VGG16, and ResNet50 on dermoscopic and macroscopic image. This research also uses black-hat filtering and inpainting at the preprocessing stage to remove hair from the skin image. Results: MobileNet is the best model for classifying melanomas or moles in this experiment which has 83.86% of F1 score and 11 second of training time per epoch. Conclusion: MobileNet and Xception have high average F1 scores of 84.42% and 80.00%, so they can detect melanoma accurately even though the number of melanoma datasets is less than moles. Therefore, it can be concluded that MobileNet and Xception are suitable models for classifying melanomas and moles. However, MobileNet has the fastest training time per epoch which is 11 seconds. In the future, oversampling method can be implemented to balance the number of datasets to improve the performance of the classification model.

A CNN-Based Model for Early Melanoma Detection

Melanoma identification and classification model based on fine-tuned convolutional neural network

MelaNet: an effective deep learning framework for melanoma detection using dermoscopic images

Automatic Classification of Melanoma Skin Cancer with Deep Convolutional Neural Networks

AI Techniques of Dermoscopy Image Analysis for the Early Detection of Skin Lesions Based on Combined CNN Features

Melanoma Skin Cancer Classification based on CNN Deep Learning Algorithms

Melanoma Detection Using Deep Learning-Based Classifications

Automated Detection of Nonmelanoma Skin Cancer Based on Deep Convolutional Neural Network

An Intelligent Mechanism to Detect Multi-Factor Skin Cancer

Melanoma Skin Cancer Detection Using Deep learning, Neural Network and Classical Machine Learning Techniques

DSCC_Net: Multi-Classification Deep Learning Models for Diagnosing of Skin Cancer Using Dermoscopic Images

Early Melanoma Detection Based on a Hybrid YOLOv5 and ResNet Technique

Automated deep learning approach for classification of malignant melanoma and benign skin lesions

Detection and Classification of Melanoma Skin Cancer Using Image Processing Technique

Melanoma Skin Cancer Identification with Explainability Utilizing Mask Guided Technique

Deep Learning-Based Classification for Melanoma Detection Using XceptionNet

Melanoma Classification Using a Novel Deep Convolutional Neural Network with Dermoscopic Images

Melanoma Detection using Convolutional Neural Network with Transfer Learning on Dermoscopic and Macroscopic Images

Melanoma Skin Cancer Detection and Classification using Deep Learning

A Deep CNN Model for Skin Cancer Detection and Classification

DSCIMABNet: A novel multi-head attention depthwise separable CNN model for skin cancer detection