Deep-MI/FastSurfer

What it solves

FastSurfer provides a fast, deep-learning-based alternative to FreeSurfer for neuroimaging analysis. It drastically reduces the time required for volumetric analysis (down to minutes) and surface-based thickness analysis (down to approximately one hour), while maintaining accuracy for brain MRI processing.

How it works

The pipeline is split into two primary components:

1. Segmentation (seg): Uses deep learning networks to perform fast segmentation and volumetric calculations. This includes specialized modules for the whole brain (asegdkt), the corpus callosum (cc), the cerebellum (cereb), and the hypothalamus (hypothal).
2. Surface Reconstruction (recon-surf): Reconstructs cortical surfaces, maps cortical labels, and performs point-wise and ROI thickness analysis. This part utilizes some FreeSurfer binaries internally.

Additionally, the FastSurfer-LIT extension allows for lesion inpainting when a lesion mask is provided, helping to handle tumors or cavities in the brain images.

Who it’s for

It is designed for researchers in neuroimaging and neuroscience who need to compute brain statistics from MR images for population data aggregation or group comparisons.

Highlights

• High Speed: Reduces volumetric analysis time to minutes and surface analysis to around one hour.
• Comprehensive Segmentation: Includes dedicated modules for the whole brain, cerebellum, hypothalamus, and corpus callosum.
• Resolution Support: Supports high-resolution images (up to 0.7mm isotropic).
• Flexible Deployment: Available via Docker, Singularity/Apptainer, macOS package, or native Linux installation.
• Lesion Handling: Experimental support for lesion inpainting via FastSurfer-LIT.