https://github.com/yousefis/Hadamard-te-ASL-recon

https://github.com/yousefis/Hadamard-te-ASL-recon

Last synced: 28 days ago
JSON representation

• Host: GitHub
• URL: https://github.com/yousefis/Hadamard-te-ASL-recon
• Owner: yousefis
• Created: 2020-05-12T09:30:43.000Z (over 4 years ago)
• Default Branch: master
• Last Pushed: 2022-04-16T18:06:15.000Z (over 2 years ago)
• Last Synced: 2024-08-03T06:01:15.123Z (4 months ago)
• Language: Python
• Size: 11.4 MB
• Stars: 7
• Watchers: 1
• Forks: 0
• Open Issues: 0
• Metadata Files:
  • Readme: README.md

Awesome Lists containing this project

• awesome_medical - Hadamard-te-ASL-recon

README

        
# Goal

In this project I designed a deep learning network to accelerate 4D data reconstruction in Tensorflow.

# Dynamic Angiography and Perfusion Reconstruction from Hadamard-te Arterial Spin Labeling of rank 8

# 1- Introduction

In this work 4D Angiography and Perfusion at eight time-points are reconstructed from an interleaved half-sampled crushed and non-crushed Hadamard-te arterial spin labeling (ASL) of rank 8. The network uses DenseUnet structure and multi-stage loss function. Different loss functions have been applied for training including: perceptual loss (PL), mean squre error (MSE), Structural Similarity Index (SSIM) in a single and multi-stage fasions. Also, a framework for generating dynamic ASL scans based on the Hadamard ASL kinetic model has been proposed. 

The reconstruction process can be formulated as: 

,

in which  is the decoding and subtraction function,  and  are the acquired scans of the  row of non-crushed and crushed Hadamard te-pCASL datasets,  and  denote perfusion and angiography scans respectively. 

Here you can find the Hadamard te-ASL signal generator.

# 2- Citation

    @inproceedings{yousefi2019fast,

    title={Fast Dynamic Perfusion and Angiography Reconstruction using an end-to-end 3D Convolutional Neural Network},

    author={Yousefi, Sahar and Hirschler, Lydiane and van der Plas, Merlijn and Elmahdy, Mohamed S and Sokooti, Hessam and Van Osch, Matthias and Staring, Marius},

    booktitle={International Workshop on Machine Learning for Medical Image Reconstruction},

    pages={25--35},

    year={2019},

    organization={Springer}

    }

# 3- Proposed network



    

    Figure 1- Proposed network, a multi-stage DenseUnet. Inputs: an interleaved half-sampled crushed and non-crushed Hadamard-te arterial spin labeling (ASL) of rank 8. Output: dynamic angiography and perdusion scans at 8 time-point.



# 4- Proposed data generator

 



    

    Figure 2- Proposed data generator.



# 5- Results



    

    Figure 3- Results of reconstructed angiography scans for one subject.





    

    Figure 4- Results of reconstructed perfusion scans for one subject.



# Requirments

Tensorflow<2 & python>3.4

# If this repository helps you in anyway, show your love :heart: by putting a :star: on this project