Convolutional Neural Network for Skin Lesion Classification: Understanding the Fundamentals Through Hands-On Learning

Area of Science:

• Dermatology
• Medical Imaging
• Computer Science

Background:

• Deep learning excels in image classification, including dermatology.
• Non-experts need accessible methods to understand deep learning for image-based diagnosis.

Purpose of the Study:

• To provide an intuitive, hands-on pedagogical activity for understanding deep learning algorithms.
• To lower technical barriers for non-experts to engage with deep learning tools.

Main Methods:

• Developed an open-source, no-installation activity for training convolutional neural networks (CNNs).
• Used a dataset of skin lesion images for different skin cancer categories.
• Provided step-by-step code descriptions and visualizations.

Main Results:

• The activity facilitates intuitive comprehension of CNNs, including convolution.
• Users can tune hyperparameters and modify code for deeper understanding.
• Realistic examples with result visualization and evaluation are included.

Conclusions:

• Hands-on experience with simplified models enhances understanding of deep learning in dermatology.
• Open-source tools reduce technical barriers, promoting broader accessibility.
• This approach supports learning without advanced computational or mathematical skills.