Automated Melanocytic Lesion Classification: Capsule Networks Trained With Synthetic Images Can Outperform Networks Trained With Real Images.

Key Points

• The study aims to evaluate the effectiveness of capsule networks trained with synthetic images for classifying melanocytic lesions.
• Utilized capsule networks (CN) for image classification
• Trained the CN on a large dataset of synthetic images of melanocytic lesions
• Compared performance against traditional convolutional neural networks (CNN)
• Capsule networks showed improved generalisation performance over CNNs
• Training with synthetic images led to better outcomes compared to using real images
• Results suggest high potential for using synthetic data in automated lesion classification

Abstract