New toolboxes, segmentation workflow and Muscle-BIDS pipeline
What's new in QMRITools 5
QMRITools 5 requires Wolfram Language 15.0 or newer. The complete list of changes, per major version, is in the changelog.
New toolboxes
- NeuralNetworkTools: building blocks for neural networks, loss layers, transfer learning and inspection of trained networks.
- ShapeTools: statistical shape models and meshes of segmented structures.
- AmaresTools: AMARES fitting of MR spectroscopy data.
Muscle segmentation
- Complete train-and-deploy workflow with new architectures (UNET++, strided convolutions), training on multiple contrasts, masked pretraining and encoder freezing for transfer learning, and parallel data loading.
- More anatomy, including the shoulder, and merging of segmentations.
- Segmentation confidence maps and QC plots.
- Command line scripts for installing QMRITools and segmenting data without Mathematica.
See automated muscle segmentation.
Muscle-BIDS
- An end-to-end pipeline from DICOM to per-muscle results, including T2 mapping, tractography and
MuscleBidsAnalysis. - Better DICOM conversion with multiple sessions and repeated scans, combining of config files and batch processing.
- Integration with MuscleMap.
- A new manual with example configurations.
See Muscle-BIDS.
Diffusion, tractography and fasciculation
- Random permeable barrier model (RPBM) fitting, correction for gradient non-linearity and export of b-values and b-vectors.
- Import and export of MRtrix
.tckfiles and tract length and curvature measures. - Area based activation sizes for fasciculation analysis.
Denoising, reconstruction, spectroscopy and simulation
- Self-supervised network denoising (
P2SDenoise), harmonic denoising of tensors and denoising of spectroscopic imaging data. - More reconstruction options: prewhitening, SENSE smoothing and weighting, k-space averaging.
SpectraSimulatorand EPG simulation that accounts for the refocusing pulse phase.
Plotting
- Dark mode,
LoessPlotand Crameri scientific colormaps (ScientificColorData).