Tutorials

Every tutorial is a self-contained folder under tutorials/ with runnable training, inference and evaluation scripts. Set the dataset paths at the top of a script, run it from inside its folder, and find checkpoints (experiments/), logs (logs/) and result tables (results/) next to it.

New to MIR?

Start with brain_registration: affine + deformable registration with pretrained weights.

Register a brain MRI

Training your own model?

lumir25_TransMorph is the most complete training template; the training guide explains it.

Training a model

Brain MRI: Learn2Reg LUMIR

Inter-subject brain MRI registration on the LUMIR challenge data (Learn2Reg 2024 and 2025). Every LUMIR25 tutorial uses the same protocol, so results are directly comparable: training with random multi-contrast augmentation, then evaluation on the LUMIR test splits and zero-shot datasets (Dice, landmark TRE, non-diffeomorphic volume).

Folder	Contents
lumir25_TransMorph, lumir24_TransMorph	TransMorph-TVF training and evaluation.
lumir25_VFA, lumir24_VFA	VFA training and inference, HyperVFA-SPR, SynthSR preprocessing, a Docker submission example.
lumir25_VoxelMorph, lumir25_LessNet, lumir25_FourierNet	The same protocol for the convolutional baselines.
lumir25_ConvexAdam, lumir24_ConvexAdam	ConvexAdam baselines (no training).
lumir25_TM_LDDMM and lumir25_*_LDDMM	LDDMM training and evaluation for TransMorph and other backbones (LDDMM).

Brain MRI: IXI and general use

Folder	Contents
brain_registration	Affine + deformable registration with pretrained weights (Register a brain MRI).
IXI_benchmarking	Benchmark seven models on IXI from a notebook (Benchmark on IXI).
IXI_TransMorphSPR, IXI_SITReg	Atlas-to-subject training with spatially varying regularization, and SITReg.
IXI_HyperTransMorph, IXI_HyperMorph	Hypernetwork-conditioned regularization.
IXI_deedsBCV	deedsBCV validation (Linux only).
build_brain_template	Population template construction (Build a brain template).
brain_MRI_preprocessing	Bias-field correction, intensity normalization, skull stripping.
pretraining_registration_DNNs	Pretraining registration networks on random synthetic images.

Whole-body PET/CT

Atlas-to-subject registration of FDG PET/CT from the autoPET challenge data, with spatially varying regularization learned or optimized.

Folder	Contents
autoPET_TransMorphSPR	TransMorph-TVF-SPR training and inference (the MedIA 2025 SPR models).
autoPET_ConvexAdamSPR, autoPET_FireANTs_SPR	Optimization-based registration with spatially varying regularization.
AutoPET	Dataset metadata summaries.

Feature learning

Folder	Contents
VFA_image_seg, VFA_image_synthesis	Segmentation and synthesis heads on pretrained VFA features.

Deployment

Folder	Contents
web_playground_HyperVFASPR_2D	Train 2D HyperVFA-SPR / HyperVFA on IXI slices, export to ONNX and run them in a web page with ONNX Runtime Web: the live playground (Registration in the browser).

Pretrained weights and dataset files used by the tutorials are listed in Pretrained weights.