SegSRGAN

SegSRGAN performs simultaneous super-resolution reconstruction and segmentation of neonatal brain MRI using generative adversarial networks (GANs) to improve cortical segmentation and image resolution from low-anisotropic clinical scans.

Key Features:

• Simultaneous reconstruction and segmentation: Integrates super-resolution reconstruction and segmentation within a single framework for neonatal brain MRI.
• Generative Adversarial Networks (GANs): Employs GANs with generator and discriminator networks to produce high-fidelity super-resolved images.
• Network architecture: Provides detailed network architecture descriptions covering generator and discriminator components and their segmentation outputs.
• Data augmentation strategies: Utilizes data augmentation during training to mitigate limited availability of high-quality medical imaging data.
• Evaluation metrics: Includes quantitative and qualitative validation using Dice coefficient, peak signal-to-noise ratio (PSNR), structural similarity (SSIM), and number of connected components.

Scientific Applications:

• Neonatal cortical analysis: Enhances cortex analysis from neonatal MR images by improving resolution and segmentation quality on low-anisotropic scans.
• Research datasets (dHCP): Applicable to processing and analysis of developing Human Connectome Project (dHCP) neonatal MRI datasets.
• Clinical datasets (Epirmex): Applicable to clinical neonatal MRI datasets such as Epirmex for improved reconstruction and segmentation.

Methodology:

Uses generative adversarial networks (generator and discriminator) to perform simultaneous super-resolution and segmentation, applies data augmentation during training, and evaluates results quantitatively (Dice coefficient, PSNR, SSIM, number of connected components) and qualitatively.