Skin Lesion Classification Using GAN based Data Augmentation.

Area of Science:

• Artificial Intelligence
• Dermatology
• Medical Imaging

Background:

• Early skin cancer detection and monitoring are vital for patient survival.
• Automated classification of dermoscopic images using deep learning shows promise for early detection.
• Limited training dataset sizes hinder the full potential of deep learning in medical image classification.

Purpose of the Study:

• To investigate the use of generative adversarial networks (GANs) for creating realistic dermoscopic images.
• To augment existing training datasets with GAN-generated images for skin lesion classification.
• To enhance the performance of deep convolutional neural networks (CNNs) in skin lesion classification.

Main Methods:

• Utilized generative adversarial networks (GANs) to synthesize realistic dermoscopic images.
• Augmented a dataset of dermoscopic images with GAN-generated examples.
• Trained and evaluated a deep convolutional neural network (CNN) on the augmented dataset.
• Compared performance gains against conventional data augmentation techniques.

Main Results:

• GAN-based data augmentation resulted in significant performance improvements for skin lesion classification.
• Synthetically generated dermoscopic images were realistic and effective for training deep learning models.
• Performance gains from GAN augmentation surpassed those from conventional methods.

Conclusions:

• Generative adversarial networks (GANs) offer a powerful approach for augmenting medical imaging datasets.
• GAN-based augmentation can significantly enhance the accuracy of deep learning models for skin cancer detection.
• This technique holds potential for improving early detection rates and patient survival in dermatology.