Tools
Open-source software I develop for MR spectroscopy. Each is on GitHub and PyPI. Contributions and feedback are welcome.
-
Augmentrum
Data augmentation package for MR spectroscopy.
A modular framework for creating diverse, physically consistent MRS datasets through flexible augmentation: k-space resampling, coil and average sampling, signal-level perturbations, and synthetic artifacts. Built on the NIfTI-MRS standard with on-the-fly augmentation for PyTorch, TensorFlow, Keras, and JAX.
-
BasisREMY
Study-specific basis set generation for MRS.
A unified framework that generates study-specific basis sets directly from raw MRS data, integrating real pulse shapes and acquisition parameters. It extracts MRSinMRS metadata across vendors, drives multiple simulation backends (FID-A, MRSCloud, FSL-MRS, and more), and exports ready-to-use basis sets with quality-control plots.
-
MIOX
Open any MR file from any vendor into one object.
A universal reader that loads MR spectroscopy, images, raw k-space, basis sets, fit results, sequences, protocols, physio recordings, and NMR data from any vendor or toolbox into one well-defined object: data, geometry, and all metadata. It keeps the complete vendor header, records where every standard key came from, and ships a quick-look viewer.
-
MRSandbox
A simulation sandbox for MR spectroscopy.
Simulates MR spectra from first principles: a virtual brain supplies the tissue at any voxel, a virtual scanner the field and hardware limits, and flexible pulse sequences probe both through a density-matrix engine. Every quantity can be fixed, sampled, or read from a map, and simulations run batched and differentiable on JAX, PyTorch, and TensorFlow.
-
MRSuite
Composable toolkit for MRS processing, analysis, and AI.
A modular framework that chains steps from any MRS toolbox (FID-A, Osprey, spant, FSL-MRS, LCModel, and more) into one pipeline, with every step recorded in the NIfTI-MRS history. It adds an AI model zoo with Hugging Face sharing, a benchmarking suite for comparing methods against ground truth, and a graphical interface that writes out its own Python.
-
NIfTI-MRS+
Batch-aware NIfTI-MRS data handling for MRS.
An extension of the NIfTI-MRS standard that wraps multiple NIfTI-MRS objects into a single batch-aware tensor structure while preserving headers, dimension tags, and provenance. Backend agnostic with NumPy, PyTorch, JAX, TensorFlow, and Keras, plus flexible indexing, sub-batch slicing, and efficient caching.
-
PyLCModel
Python wrapper for LCModel spectral fitting in MRS.
A lightweight Python wrapper that streamlines least-squares spectral fitting with LCModel. It automates control-file generation, resolves the LCModel binary for you, accepts flexible inputs (NumPy, NIfTI-MRS, jMRUI,
.RAW), and parses the full output with single- or multi-core batch fitting. -
PyTARQUIN
TARQUIN fitting, straight from Python.
Run TARQUIN without installing it: the binary is fetched or run in a container automatically, with builds verified against the official release. Use any TARQUIN option or its internal basis simulated at your echo time, and get every reported result back in Python.