diagnostics4u

LiverUSRecon: Automatic 3D Reconstruction and Volumetry of the Liver with a Few Partial Ultrasound Scans

MICCAI 2024 Paper (arXiv) Code

Authors

Kaushalya Sivayogaraj, Sahan T. Guruge, Udari Liyanage, Jeevani Udupihille, Saroj Jayasinghe, Gerard Fernando, Ranga Rodrigo, Rukshani Liyanaarachchi

📰 News

Code is available at LiverUSRecon.
To request access to the dataset, please contact Kaushalya Sivayogaraj.
Pre-trained weights are available:
- Segmentation model
- Reconstruction model

Abstract

LiverUSRecon Overview

3D reconstruction of the liver for volume measurement and 3D visual shape analysis using an accessible medical imaging modality like ultrasound (US) imaging is important. We present the first method capable of reconstructing the liver from a few partial ultrasound scans acquired at the midline, midclavicular line, and anterior-axillary line. To the best of our knowledge, this is the first automated deep learning method that calculates the liver volume from three incomplete 2D US scans. Further, we introduce a new US liver database with parallel, annotated CT scans comprising 134 scans. Our volumetry results are statistically closer to the ground-truth volumes obtained from CT scans than the volumes computed by radiologists using the Childs’ method.

Results

Ultrasound Segmentation and 3D Reconstruction

Overall Framework — 3D Reconstruction

3D Reconstruction Overlap

Overlap between Ground Truth and Prediction

Point-to-Point Distance

Absolute Point-to-Point Distance

Statistical Analysis

Main Results

Volume Comparison

Getting Started

1. Download Pre-trained ViT Model

Download the R50-ViT-B_16 model from Google.
Place the downloaded model in ./model/vit_checkpoint/imagenet21k/ and rename it to R50-ViT-B_16.npz.
Download the pre-trained segmentation and reconstruction models:
- Segmentation model
- Reconstruction model
Place both models in a folder named models under the results directory.

2. Prepare the Dataset

Contact Kaushalya Sivayogaraj to request access to the inference datasets.

3. Download Liver SSM Information

Download the following Statistical Shape Model (SSM) files and place them in ./SSM/:

File	Link
Shape parameters	VT.txt
Mean shape	liver_aver.obj
PCA ratio	pca_ratio.txt
Normalization info	nor_list.txt

4. Environment Setup

Create a Python 3.7 environment and install the required dependencies:

pip install -r requirements.txt

5. Inference

Run the inference script on the downloaded dataset:

CUDA_VISIBLE_DEVICES=0 python inference_liverusrecon.py \
  --inference {dataset path} \
  --save {results path} \
  --ssm_info {ssm_info path}

Licenses

Code

Code is licensed under the GNU Affero General Public License v3.0. You should have received a copy of the GNU Affero General Public License along with this code. If not, see https://www.gnu.org/licenses/.

ML Weights

ML Weights are licensed under the Creative Commons Attribution-NonCommercial-NoDerivs 3.0 Unported License. You should have received a copy of the license along with this work. If not, see https://creativecommons.org/licenses/by-nc-nd/3.0/.

Patient Data

Patient data is licensed under the Creative Commons Attribution-NonCommercial-NoDerivs 3.0 Unported License. You should have received a copy of the license along with this work. If not, see https://creativecommons.org/licenses/by-nc-nd/3.0/.

Citation

If you find this work useful, please consider citing:

@InProceedings{Siv_LiverUSRecon_MICCAI2024,
    author    = {Sivayogaraj, Kaushalya and Guruge, Sahan I. T. and Liyanage, Udari A. and
                 Udupihille, Jeevani J. and Jayasinghe, Saroj and Fernando, Gerard M. X. and
                 Rodrigo, Ranga and Liyanaarachchi, Rukshani},
    title     = ,
    booktitle = {Proceedings of Medical Image Computing and Computer Assisted Intervention},
    year      = {2024},
    pages     = {436--445}
}

This site is open source. Improve this page.