Melanoma classification using generative adversarial network and proximal policy optimization.

Area of Science:

• Oncology
• Artificial Intelligence
• Medical Imaging

Background:

• Melanoma is a serious cancer often caused by UV radiation, necessitating early detection due to its aggressive nature.
• Current classification methods face challenges with imbalanced datasets and achieving high accuracy.

Purpose of the Study:

• To develop and validate a novel deep learning framework for enhanced melanoma classification.
• To address data imbalance and improve model generalizability in melanoma detection.

Main Methods:

• Utilized a deep learning framework with three dilated convolution layers for feature extraction.
• Incorporated Off-policy Proximal Policy Optimization (PPO) to manage imbalanced training data.
• Employed a Generative Adversarial Network (GAN) with novel regularization for data augmentation and stable training.

Main Results:

• Achieved an F-measure of 91.836% and a geometric mean of 91.920%.
• Demonstrated superior performance compared to existing models.
• Validated on the SIIM-ISIC Melanoma Classification Challenge-ISIC-2020 dataset.

Conclusions:

• The proposed framework shows significant potential for improving early melanoma detection.
• The model's performance suggests practical utility in clinical settings for more accurate treatment planning.
• Advances in AI-driven diagnostics can significantly aid in combating aggressive cancers like melanoma.