MOOSE (Multi-organ objective segmentation) is a data-centric AI solution for multi-organ segmentation in whole-body PET/CT images, targeting researchers in systemic TB and related fields. It leverages a nn-UNet-based pipeline to segment up to 120 tissue classes, offering significant speed and efficiency improvements over previous versions.

How It Works

MOOSE 3.0 utilizes a nn-UNet architecture, optimized for speed and memory efficiency. It employs Dask for in-memory processing, avoiding disk writes and enabling efficient handling of large datasets on standard hardware. The system supports multi-instance parallelization ("Herd Mode") for scaling inference across multiple compute resources.

Quick Start & Requirements

• Install: pip install moosez
• Prerequisites: Python 3.10, 16GB RAM. NVIDIA GPU recommended for acceleration, but CPU and Apple Silicon (MPS) are supported.
• Setup: Installation is straightforward via pip.
• Docs: Usage Guide

Highlighted Details

• 3x faster than MOOSE 2.0.
• Runs natively on Apple Silicon (MPS) for GPU-like speeds.
• Supports batch processing via CLI and programmatic use as a Python library.
• Offers 10 distinct segmentation models for clinical and preclinical imaging.

Maintenance & Community

• Active development with contributions from multiple individuals and Kitware, Inc.
• Project is part of the enhance.pet community.

Licensing & Compatibility

• The README does not explicitly state a license. Compatibility for commercial use or closed-source linking is not specified.

Limitations & Caveats

• Requires specific directory structures and NIFTI file naming conventions for batch processing; non-compliance leads to skipped subjects.
• The README does not specify a license, which may impact commercial adoption.