ROCKETSHIP
Dynamic contrast-enhanced MRI analysis software for parametric
and DCE workflows.
- Focus
- Parametric MRI + DCE-MRI processing
- Environment
- MATLAB-based workflows
- Acceleration
- Optional NVIDIA GPU via Gpufit
- License
- GPL-2.0
-
Processes parametric MRI and DCE-MRI files, originally developed at
the Biological Imaging Center at Caltech.
-
Designed to keep data prep, model fitting, and review steps in a
single workflow.
Reference:
Barnes et al., “ROCKETSHIP: a flexible and modular
software tool for the planning, processing and analysis of dynamic
contrast-enhanced MRI studies,” BMC Medical Imaging (2015).
Gpufit
GPU-accelerated curve fitting with Levenberg-Marquardt optimization,
adapted for MRI fitting workflows.
- Focus
- CUDA accelerated Levenberg-Marquardt curve fitting
- Environment
- CUDAs
- Bindings
- Python, MATLAB, Java
- License
- MIT
-
Fork with updated CMake build system for modern compilers and Linux compatibility.
-
Approximately 40x faster than the compiled CPU-only implementation
and >1000x faster than interpreted MATLAB-based alternatives.
-
Added support for bounded fitting and numerical differentiation.
Reference:
Przybylski et al., “Gpufit: An open-source toolkit for
GPU-accelerated curve fitting,” Scientific Reports (2017).
Vascular Function
Auto AIF detection for DCE-MRI data and ROCKETSHIP using a neural network.
- Focus
- Automatic AIF detection for DCE-MRI processing
- Environment
- Keras/TensorFlow
- Model
- UNet with pretrained weights
- License
- MIT
-
Keras/TensorFlow implementation for automatic AIF detection in DCE-MRI data
and ROCKETSHIP workflows.
-
Inference predicts an AIF curve and 3D mask with
automated normalization and resampling steps.
-
Includes training and inference scripts plus a pretrained model.
Reference:
Saca et al., “Automatic detection of arterial input function for brain
DCE-MRI in multi-site cohorts,” Magnetic Resonance in Medicine (2025).
Collaborative Projects
AT Seg 3D CNN
Automated abdominal adipose tissue segmentation and volume
quantification on longitudinal MRI using multi-contrast 3D CNNs.
- Focus
- Abdominal SAT/VAT segmentation
- Environment
- PyTorch
- Models
- ACD 3D U-Net + 3D nnU-Net
- License
- Academic Software License
-
Full-volume multi-contrast inputs stack opposed-phase, water,
and fat images as channels for volumetric segmentation.
-
Outputs SAT, VAT, and background masks for adipose tissue
quantification.
-
Includes preprocessing and training/inference scripts for the
3D CNN pipelines.
Reference:
Kafali et al., “Automated abdominal adipose tissue segmentation and
volume quantification on longitudinal MRI using 3D convolutional
neural networks with multi-contrast inputs,” Magnetic Resonance
Materials in Physics, Biology and Medicine (2024).
FL Ped AT Seg
Federated learning framework for pediatric abdominal VAT/SAT
segmentation on free-breathing 3D Dixon MRI.
- Focus
- Pediatric VAT/SAT segmentation + quantification
- Environment
- PyTorch
- Model
- 3D U-Net with Cross-cohort federated learning
- License
- Academic Software License
-
Trains across pediatric and adult cohorts without direct data
sharing, retaining local decoders for domain specificity.
-
Produces VAT and SAT masks for volumetric quantification and PDFF
comparison.
-
Achieves strong Dice scores and rapid inference for pediatric
abdominal adipose tissue segmentation.
Reference:
Wu et al., “Cross-cohort federated learning for pediatric abdominal
adipose tissue segmentation on free-breathing 3D Dixon MRI,”
Radiology Advances (2025).