Skin Cancer Detection Based on Transfer Learning Techniques

Skin cancer detection and classification are critical for early diagnosis and treatment, particularly with melanoma contributing significantly to global mortality rates. This paper presents a study that uses transfer learning with convolutional neural networks (CNNs) to classify skin cancer images as Benign or Malignant. Two pre-trained CNN architectures: InceptionV3, and Xception, are employed with transfer learning and fine-tuning and evaluated for their performance. Notably, InceptionV3 achieved a precision of 0.8914, recall of 0.8909, F1 score of 0.8910, accuracy of 0.8909, and ROC/AUC of 0.8911. These metrics, along with those from the Xception model, are compared to assess their efficacy in early detection. InceptionV3 outperforms state-of-the-art methods, demonstrating its superiority in skin cancer classification. The study aims to contribute to dermatological healthcare by enhancing skin cancer diagnosis through efficient image classification techniques.