This is a repository for the project "An End-to-End Pipeline for Automatic Patient-Specific Cranial Implant Design: From CT Scans to Titanium Implants". We aim to make the repository as an easy starter for performing Autoimplant. Please see our paper "CraNeXt: Automatic Reconstruction of Skull Implants With Skull Categorization Technique" and model at guitared/Cranext if you want to see recent development from the lab.
The repository contains the following implementations:
- Automatic skull segmentation predicts 3D skulls from a given CT Scans in grayscale
- Autoimplant predicts complete skull from a defective skull
- Autoimplantpipe contains a library for autoimplant pipeline described in the paper.
- Notebooks folder contains notebook for automatic skull segmentation and autoimplant inferences including
01_autosegmentation.ipynbis an example notebook for performing skull segmentation02_autoimplant_prediction.ipynbis an example notebook for performing autoimplant inference
- Scripts folder contains scripts for training automatic skull segmentation and autoimplant models.
We provide the pretrained models for segmentation and autoimplant below. Additionally, we provide an example SkullBreak data for autoimplant inference data folder.
Dice score and model checkpoints for segmentation and autoimplant models.
| Segmentation model | Dice Score | Checkpoint |
|---|---|---|
| Unet + Post Process | 0.9100 | link |
| Autoimplant model | Dice Score | Checkpoint |
|---|---|---|
| PCA | 0.2207 | notebook |
| 3DUNetCNN off-the-shelf | 0.0924 | link |
| 3DUNetCNN SkullBreak | 0.6031 | link |
| 3DUNetCNN in-house | 0.7881 | link |
| 3DUNetCNN SkullBreak + in-house | 0.7936 | link |
Note: Model checkpoints trained with in-house dataset is distributed under CC BY-NC license. The source code is distributed under Apache 2.0.
Download the repository using git:
git clone https://github.com/biodatlab/autoimplantpipe
cd autoimplantpipeInstall dependencies and library using pip:
pip install -r requirements.txt # install dependencies
pip install . # install `autoimplantpipe` library