voxelmorph: a learning-based framework for deformable medical image registration and alignment

What it solves

VoxelMorph provides a learning-based framework for image registration, which is the process of aligning two images (such as medical scans) by modelling deformations. It addresses the challenge of aligning images with different anatomical structures or different imaging modalities (e.g., CT to MRI) efficiently.

How it works

The library uses convolutional neural networks (CNNs) to predict deformation fields that warp one image (the moving image) to match another (the fixed image or an atlas). It supports several registration types, including unsupervised learning with various loss functions (MSE, CC), diffeomorphic registration (which ensures the deformation is smooth and invertible), and affine registration. It also includes SynthMorph, a strategy for training registration models using synthesized images rather than real acquired imaging data, making the models contrast-invariant.

Who it’s for

It is designed for researchers and practitioners in medical imaging and computer vision who need to align 3D volumes or images and model spatial deformations.

Highlights

• Multi-modal Support: Capable of performing registration across different imaging modalities, such as CT-to-MRI.
• SynthMorph: Allows training of registration networks without the need for real training data by using synthesized images.
• Diffeomorphic Mapping: Supports the creation of smooth, invertible deformations to maintain anatomical topology.
• ** uma TensorFlow and PyTorch**: Offers stable TensorFlow implementations and an active PyTorch development branch.
• Atlas Construction: Includes tools for creating unconditional and conditional deformable templates (atlases).