Big transfer learning for automated skin cancer classification

<p>Skin cancer is an example of the most dangerous disease. Early diagnosis skin can save many people’s lives. Manual classification methods are time-consuming and costly. Deep learning has been proposed for automated cancer. Although deep showed impressive performance in several medical imaging tasks, it requires a big number images to achieve good performance. The task suffers from providing with sufficient data due expensive annotation process required experts. One used solutions transfer pre-trained models ImageNet dataset. However, learned features different image features. To end this, we introduce novel approach by training (VGG, GoogleNet, ResNet50) on large unlabelled images, first. We then train them small labeled images. Our experimental results proved that method efficient achieving accuracy 84% ResNet50 when directly trained 93.7% approach.</p>