Scholarlink

<p>Melanoma is a life-threatening skin cancer that requires early and accurate diagnosis to improve patient outcomes. Melanoma presents substantial problems in early identification and diagnosis, and it can be fatal if not discovered promptly. While an efficient deep neural network requires numerous annotated images, real-world settings like medical image data, are frequently scarce. To address this issue, pre-trained models such as InceptionV3 can benefit from the transfer learning approaches and image augmentation. Therefore, this study proposes an enhanced convolutional neural network (CNN) architecture that integrates transfer learning with progressive layer unfreezing and domain-specific data augmentation to address challenges of limited labeled data and class imbalance. Using the Kaggle dermoscopic dataset, the model was fine-tuned on an InceptionV3 backbone and compared with DenseNet121, VGG16, and a custom CNN. The VGG16-based transfer learning model achieved the highest performance with 97% accuracy, 0.95 F1-score, and 0.95 ROC-AUC, outperforming baseline CNNs and conventional transfer learning approaches. The results demonstrate the robustness and practical potential of the proposed framework for real-world melanoma detection applications.</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

An Advanced Convolutional Neural Network Architecture Utilizing Transfer Learning for Melanoma Detection

  • 期刊论文

  • 发表日期：

    2025年12月12日

查看全文
  • Joseph Bamidele Awotunde,
  • Korede Israel Adeyanju,
  • Kehinde Elisha Akerele,
  • Oluwatobi Akinlade,
  • Sakinat Oluwabukonla Folorunso,
  • Sunday Adeola Ajagbe

摘要

Melanoma is a life-threatening skin cancer that requires early and accurate diagnosis to improve patient outcomes. Melanoma presents substantial problems in early identification and diagnosis, and it can be fatal if not discovered promptly. While an efficient deep neural network requires numerous annotated images, real-world settings like medical image data, are frequently scarce. To address this issue, pre-trained models such as InceptionV3 can benefit from the transfer learning approaches and image augmentation. Therefore, this study proposes an enhanced convolutional neural network (CNN) architecture that integrates transfer learning with progressive layer unfreezing and domain-specific data augmentation to address challenges of limited labeled data and class imbalance. Using the Kaggle dermoscopic dataset, the model was fine-tuned on an InceptionV3 backbone and compared with DenseNet121, VGG16, and a custom CNN. The VGG16-based transfer learning model achieved the highest performance with 97% accuracy, 0.95 F1-score, and 0.95 ROC-AUC, outperforming baseline CNNs and conventional transfer learning approaches. The results demonstrate the robustness and practical potential of the proposed framework for real-world melanoma detection applications.