kibromberihu/mipsegmentator repository overview

A 3D lesion segmentation on whole-body PET images including automated quality control.

For details refer to https://github.com/KibromBerihu/MIPsegmentatorV1/tree/main⁠

Main steps:

1. Segmentation of lesion areas on 2d MIP PET images using AI4elife (https://github.com/KibromBerihu/ai4elife⁠)
2. Reconstructs 3D segmentation from the 2D MIP segmentations
3. Generates confidence score (quality control (QC)) for each cases in excel file
4. Automatically calculates some image-based biomarkers, sTMTV, sDmax (DOI https://doi.org/10.2967/jnumed.121.263501⁠) and saves them in excel file

The model was trained mainly from FDG-PET lymphoma data.

How to run it?

Tested on linux:

1. Make sure you put the file structures according to the readme.md file:

   --- input_output_data

   --- input             
   
   --- output
   

   Copy

2. docker pull kibromberihu/mipsegmentator:latest-0

3. docker run --gpus all --ipc=host --rm -v "/path/to/input_output/input/":"/home/docker_input" -v "/path/to/input_output/output/":"/home/docker_output" kibromberihu/mipsegmentator:latest-0

All output files will be saved under the folder '/path/to/input_output_data/output/predicted_data"

Please consider citing the following papers:

1. DOI: https://doi.org/10.2967/jnumed.121.263501⁠

2. DOI: https://doi.org/10.1109/TMI.2021.3060497⁠

3. DOI: https://doi.org/10.1038/s41597-022-01718-3⁠