Scholarlink

<p>Skin cancer is a widespread and potentially fatal disease that requires accurate diagnosis for effective treatment. A novel method for improving the accuracy of skin cancer detection and classification using a Convolutional Neural Network (CNN) architecture is presented. The proposed approach addresses critical challenges in skin cancer recognition, including lesion segmentation, feature extraction, and merging. The method begins with optimized K-Nearest Neighbors (KNN) for skin lesion segmentation, followed by feature extraction using two advanced deep learning models: Darknet-53 and MobileNet-V2. These extracted features are reduced via an average pooling layer and then fused to classify the lesion into seven categories using a multi-class Support Vector Machine (SVM). The Peak Signal-to-Noise Ratio (PSNR) and Structural Similarity Index (SSIM) are used to evaluate the segmentation performance. In contrast, classification performance was measured using sensitivity, accuracy, specificity, F1 Score, and precision. The proposed method achieved a PSNR of 19.5224 and an SSIM of 0.9754, yielding segmentation and classification metrics of 99.3% accuracy, 85.55% sensitivity, 99.63% specificity, 91.5% precision, and an F1-score of 87.57. Finally, the proposed model outperforms single-learning models in feature extraction and classification.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Explainable AI and Optimized KNN with Hybrid Deep Learning Models for Precise Classification of Multi-Class Skin Lesions

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年5月15日

查看全文
  • Yousef S. Alsahafi,
  • Mohamed A. Kassem,
  • Khalid M. Hosny

摘要

Skin cancer is a widespread and potentially fatal disease that requires accurate diagnosis for effective treatment. A novel method for improving the accuracy of skin cancer detection and classification using a Convolutional Neural Network (CNN) architecture is presented. The proposed approach addresses critical challenges in skin cancer recognition, including lesion segmentation, feature extraction, and merging. The method begins with optimized K-Nearest Neighbors (KNN) for skin lesion segmentation, followed by feature extraction using two advanced deep learning models: Darknet-53 and MobileNet-V2. These extracted features are reduced via an average pooling layer and then fused to classify the lesion into seven categories using a multi-class Support Vector Machine (SVM). The Peak Signal-to-Noise Ratio (PSNR) and Structural Similarity Index (SSIM) are used to evaluate the segmentation performance. In contrast, classification performance was measured using sensitivity, accuracy, specificity, F1 Score, and precision. The proposed method achieved a PSNR of 19.5224 and an SSIM of 0.9754, yielding segmentation and classification metrics of 99.3% accuracy, 85.55% sensitivity, 99.63% specificity, 91.5% precision, and an F1-score of 87.57. Finally, the proposed model outperforms single-learning models in feature extraction and classification.