Generative Modelling of 3D in-silico Spongiosa with Controllable Micro-Structural Parameters

Code, data, and trained models from the MICCAI 2020 paper. We developed a method to create micro-structural bone samples in-silico that 1) contain realistic structures, and 2) allow to steer the properties to simulate changes of micro-structural parameters from deterioration or medical bone treatment.

Requirements

Python 3.+ and PyTorch 1.6.0.

The code has been tested only with PyTorch 1.6.0, there are no guarantees that it is compatible with older versions.

Clone repo

$ git clone https://github.com/emmanueliarussi/generative3DSpongiosa.git
$ cd generative3DSpongiosa

Basic usage

Before training, download paper's data and unzip it inside data folder.

$ conda env create -f environment.yml
$ conda activate generativespongiosa
$ cd code
$ python train.py

Sampling random bone

$ cd code
$ python synthesize_random_samples.py --num_samples 20

Optimizing for micro-structural parameters

$ cd code
$ python optimize_sample.py 

References

coming soon!