Mohamed Jalaleddin, Tezel Necmi Serkan, Rahebi Javad, Ghadami Raheleh
Electrical and Electronics Engineering Department, Karabuk University, 78050 Karabuk, Türkiye.
Department of Software Engineering, Istanbul Topkapi University, 34662 Istanbul, Türkiye.
Diagnostics (Basel). 2025 Mar 18;15(6):761. doi: 10.3390/diagnostics15060761.
Melanoma is a highly aggressive form of skin cancer, necessitating early and accurate detection for effective treatment. This study aims to develop a novel classification system for melanoma detection that integrates Convolutional Neural Networks (CNNs) for feature extraction and the Aquila Optimizer (AO) for feature dimension reduction, improving both computational efficiency and classification accuracy. The proposed method utilized CNNs to extract features from melanoma images, while the AO was employed to reduce feature dimensionality, enhancing the performance of the model. The effectiveness of this hybrid approach was evaluated on three publicly available datasets: ISIC 2019, ISBI 2016, and ISBI 2017. For the ISIC 2019 dataset, the model achieved 97.46% sensitivity, 98.89% specificity, 98.42% accuracy, 97.91% precision, 97.68% F1-score, and 99.12% AUC-ROC. On the ISBI 2016 dataset, it reached 98.45% sensitivity, 98.24% specificity, 97.22% accuracy, 97.84% precision, 97.62% F1-score, and 98.97% AUC-ROC. For ISBI 2017, the results were 98.44% sensitivity, 98.86% specificity, 97.96% accuracy, 98.12% precision, 97.88% F1-score, and 99.03% AUC-ROC. The proposed method outperforms existing advanced techniques, with a 4.2% higher accuracy, a 6.2% improvement in sensitivity, and a 5.8% increase in specificity. Additionally, the AO reduced computational complexity by up to 37.5%. The deep learning-Aquila Optimizer (DL-AO) framework offers a highly efficient and accurate approach for melanoma detection, making it suitable for deployment in resource-constrained environments such as mobile and edge computing platforms. The integration of DL with metaheuristic optimization significantly enhances accuracy, robustness, and computational efficiency in melanoma detection.
黑色素瘤是一种极具侵袭性的皮肤癌，需要早期准确检测以进行有效治疗。本研究旨在开发一种用于黑色素瘤检测的新型分类系统，该系统集成了用于特征提取的卷积神经网络（CNN）和用于特征降维的天鹰座优化器（AO），以提高计算效率和分类准确率。所提出的方法利用CNN从黑色素瘤图像中提取特征，同时使用AO降低特征维度，从而提升模型性能。在三个公开可用的数据集上评估了这种混合方法的有效性：ISIC 2019、ISBI 2016和ISBI 2017。对于ISIC 2019数据集，该模型实现了97.46%的灵敏度、98.89%的特异性、98.42%的准确率、97.91%的精确率、97.68%的F1分数和99.12%的AUC-ROC。在ISBI 2016数据集上，其灵敏度达到98.45%，特异性为98.24%，准确率为97.22%，精确率为97.84%，F1分数为97.62%，AUC-ROC为98.97%。对于ISBI 2017，结果为灵敏度98.44%，特异性98.86%，准确率97.96%，精确率98.12%，F1分数97.88%，AUC-ROC为99.03%。所提出的方法优于现有的先进技术，准确率高出4.2%，灵敏度提高6.2%，特异性增加5.8%。此外，AO将计算复杂度降低了高达37.5%。深度学习-天鹰座优化器（DL-AO）框架为黑色素瘤检测提供了一种高效且准确的方法，使其适用于在移动和边缘计算平台等资源受限的环境中部署。深度学习与元启发式优化的集成显著提高了黑色素瘤检测的准确率、鲁棒性和计算效率。
