PSMA-Hornet: Fully-automated, multi-target segmentation of healthy organs in PSMA PET/CT images

This repository contains an AI model and accompanying Python scripts for fully-automated simultaneous segmentation of 14 healthy organs with high tracer uptake onPET/CT images. The imaging tracer used in model development was [18F]DCFPyL.

Required model inputs: (a) DICOM PET series acquired with a PSMA tracer (tracers other than [18F]DCFPyL may work); (b) DICOM CT series The PET and CT inputs must be co-registered.

For details regarding model architecture, training methodology, and testing results please refer to the following publication: Klyuzhin IS, Chaussé G, Bloise I, et al. PSMA-Hornet: Fully-automated, multi-target segmentation of healthy organs in PSMA PET/CT images. Med Phys. 2024; 51: 1203–1216. https://doi.org/10.1002/mp.16658 PMID: 37544015

---

Requirements

• Python 3.8+
• Python packages:
  • tensorflow==2.3.1
  • numpy
  • pydicom
  • glob2
  • scipy
  • rt_utils
  • opencv-python

Installation

• You should be able to run the test inference script (main.py) from a native or virtual Python/Conda environment
• Ensure that hornetlib.py is included in your Python search path
• Ensure that unet2D_pretrain_inputfusion_vlarge_multioutput.200ep.fold0.h5 is included in your Python search path

Configuration

In the script main.py, set the variable INPUT_DIR to the folders where the PET and CT series are located

Usage (inference)

1. Copy DICOM PET and CT series in the pre-specified directory
2. Run main.py

Note: Inference results will be saved as numpy ndarray in the variables predicted_image (multi-channel) and predicted_image_lab (single-channel). You will need to convert it to the desired format.

Project Structure

psma-hornet/
├── main.py                # Python script for evaluating the model
├── hornetlib.py           # Utility functions and architecture definitions
├── hornet_model.h5        # Model trained parameters
├── README.md              # Project documentation
└── LICENSE                # License file

Licence

This project is licenced under the MIT License.

How to cite

If you are including PSMA-Hornet into your projects, kindly include the following citation:

Klyuzhin IS, Chaussé G, Bloise I, et al. PSMA-Hornet: Fully-automated, multi-target segmentation of healthy organs in PSMA PET/CT images. Med Phys. 2024; 51: 1203–1216.

Acknowledgments

This project was supported by the Canadian Institutes of Health Research Project grant PJT-162216, National Institutes of Health / Canadian Institutes of Health Research QIN grant 137993, and Mitacs Accelerate grant IT18063. Azure Cloud compute credits were provided by Microsoft for Health.