In this work, we seek to validate and segment femur images obtained from the Photon-counting computed tomography (PCCT) scanner. Using the established high-resolution peripheral quantitative computed tomography (HR-pQCT) im- ages as a ground truth. We use a U-Net architecture to segment the trabecular and cortical bone compartments. Further, introduce an auxiliary task to the U- Net architecture, to predict the HR-pQCT images from the PCCT images. We find that the U-Net architecture can segment the trabecular and cortical bone with a high accuracy. Our findings indicate that the auxiliary task does not im- prove the segmentation of the trabecular and cortical bone. We conclude that the PCCT images can be used to extract valid cortical and trabecular bone regions in an automated fashion
Create a new conda environment using python 3.11
conda create -n femur python=3.11
Activate the environment and install the required packages:
pip install -r ./requirements.txt
depending on your Cuda version some changes might be necessary.
We provide model weights here.
Then the full model pipeline can be run using the inference.py script.
With the environment active use python scripts/inference.py --image_path PATH_TO_IMAGE_FILE --model_path PATH_TO_MODEL_PTH --model_type [3D, 2D, RETINA]