Alruwaili Madallah, Mohamed Mahmood
Department of Computer Engineering and Networks, College of Computer and Information Sciences, Jouf University, Sakaka 72388, Aljouf, Saudi Arabia.
Department of Information Systems and Technology, Faculty of Graduate Studies for Statistical Research, Cairo University, Giza 12613, Egypt.
Diagnostics (Basel). 2025 Feb 25;15(5):551. doi: 10.3390/diagnostics15050551.
Medical diagnosis for skin diseases, including leukemia, early skin cancer, benign neoplasms, and alternative disorders, becomes difficult because of external variations among groups of patients. A research goal is to create a fusion-level deep learning model that improves stability and skin disease classification performance. The model design merges three convolutional neural networks (CNNs): EfficientNet-B0, EfficientNet-B2, and ResNet50, which operate independently under distinct branches. The neural network model uses its capability to extract detailed features from multiple strong architectures to reach accurate results along with tight classification precision. A fusion mechanism completes its operation by transmitting extracted features to dense and dropout layers for generalization and reduced dimensionality. Analyses for this research utilized the 27,153-image Kaggle Skin Diseases Image Dataset, which distributed testing materials into training (80%), validation (10%), and testing (10%) portions for ten skin disorder classes. Evaluation of the proposed model revealed 99.14% accuracy together with excellent precision, recall, and F1-score metrics. The proposed deep learning approach demonstrates strong potential as a starting point for dermatological diagnosis automation since it shows promise for clinical use in skin disease classification.
由于患者群体之间存在外部差异，对包括白血病、早期皮肤癌、良性肿瘤和其他病症在内的皮肤疾病进行医学诊断变得困难。一个研究目标是创建一个融合级深度学习模型，以提高稳定性和皮肤疾病分类性能。该模型设计合并了三个卷积神经网络（CNN）：EfficientNet-B0、EfficientNet-B2和ResNet50，它们在不同分支下独立运行。神经网络模型利用其从多个强大架构中提取详细特征的能力，以达到准确的结果并保持紧密的分类精度。一种融合机制通过将提取的特征传输到全连接层和随机失活层来完成其操作，以实现泛化和降维。本研究分析使用了包含27153张图像的Kaggle皮肤疾病图像数据集，该数据集将测试材料分为训练（80%）、验证（10%）和测试（10%）三部分，用于十种皮肤疾病类别。对所提出模型的评估显示准确率为99.14%，同时具有出色的精确率、召回率和F1分数指标。所提出的深度学习方法显示出作为皮肤病诊断自动化起点的强大潜力，因为它在皮肤疾病分类的临床应用中显示出前景。
