NiftyPET is an open-source Python package for high-throughput Positron Emission Tomography (PET) image reconstruction, manipulation, processing, and quantitative analysis, primarily developed at University College London (UCL).
The platform is specifically engineered for high quantitative accuracy and precision, with a strong emphasis on data acquired using hybrid and simultaneous PET/MR scanners. Its core routines are written in CUDA C and embedded in Python C extensions, enabling efficient, high-throughput execution on NVIDIA Graphics Processing Units (GPUs).
NiftyPET covers the entire processing pipeline, from raw list-mode (LM) PET data through to the final image statistic of interest (e.g., regional SUV). Key capabilities include:
- High-Fidelity Correction Models: Accurate attenuation coefficient map generation, fully 3D scatter modelling, and estimation of reduced-variance random events.
- Image Processing: Voxel-based Partial Volume Correction (PVC), image manipulation, processing, and registration (via the
nimpasub-package). - Uncertainty Estimation: Facilitates voxel-wise estimation of uncertainties of image statistics by enabling LM bootstrapping and multiple independent reconstructions.
While a key application is brain imaging in neurodegeneration (e.g., with amyloid tracers), the software is equally capable for whole-body imaging. It is a powerful tool for scientific research, allowing for the development and validation of new reconstruction and analysis algorithms within a high-performance, open-source framework.