Generative Adversarial Networks to Improve Fetal Brain Fine-Grained Plane Classification

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Biomedical Engineering

Background:

• Generative adversarial networks (GANs) are increasingly used in medical imaging across modalities like MRI and CT.
• Applications of GANs for data augmentation in ultrasound imaging, particularly for fetal brain analysis, remain less explored.
• Accurate classification of fetal brain ultrasound planes is crucial for prenatal diagnostics.

Purpose of the Study:

• To investigate the efficacy of GANs in generating synthetic fetal brain ultrasound images.
• To evaluate the performance of GAN-based data augmentation in improving fetal brain ultrasound plane classification.
• To compare GAN-augmented classifiers with baseline models and traditional augmentation techniques.

Main Methods:

• Utilized state-of-the-art Generative Adversarial Networks (StyleGAN2-ADA) for synthetic fetal brain ultrasound image generation.
• Implemented GAN-based data augmentation strategies for training classification models.
• Compared the performance of GAN-augmented classifiers against baseline classifiers using classical augmentation methods.

Main Results:

• GAN-generated data, when combined with classical augmentation, significantly improved classification accuracy.
• The area under the curve (AUC) score was enhanced by incorporating GAN-based synthetic data.
• The study demonstrated the potential of GANs to augment limited ultrasound datasets.

Conclusions:

• Generative adversarial networks show promise as a data augmentation technique in fetal brain ultrasound imaging.
• Combining GANs with traditional augmentation strategies offers a viable method to enhance classification performance.
• This approach can help overcome data scarcity challenges in medical imaging applications.