Adaptive-medical-image-deep-color-perception-algorithm

"Adaptive medical image deep color perception algorithm"

Setup

This code is based on tensorflow. It has been tested on Ubuntu 14.04 LTS.

Dependencies:

• Tensorflow 1.40
• Matlab R2016b
• Imagehash
• Numpy
• Pillow
• PyCUDA

It is recommended to use Anaconda Python, since you only need to install Tensorflow and PyCUDA manually to setup. The CUDA is optional but really recommended

CUDA backend:

• CUDA
• cudnn

VGG-19 model weights

The VGG-19 model of tensorflow is adopted from VGG Tensorflow with few modifications on the class interface. The VGG-19 model weights is stored as .npy file and could be download from Google Drive or BaiduYun Pan. After downloading, you need to fine tune the weight of VGG19 by medical images. Then copy the fine-tuned weight file to the ./vgg19 directory.

Usage

Basic Usage

You need to specift the path of the content image and then run the command

python Adaptive_MIdeepcolor.py --content_image_path <path_to_content_image> --color_option 2

Example:

python Adaptive_MIdeepcolor.py --content_image_path ./ct.jpg --color_option 2

Other Options

--color_option specifies three different ways of medical image colorization. --color_option 0 is to generate colored medical image without Y-loss. This result is similary to artistic work, its texture details can't be preserved. --color_option 1 is to generate colored medical image with Y-loss, its texture details is preserved. --color_option 2 is to set the generated result of --color_option 0 as the initial image of --color_option 1, then use Y-loss to constrain texture details.

--content_weight specifies the weight of the content loss (default=1), --color_weight specifies the weight of the style loss (default=100), --tv_weight specifies the weight of variational loss (default=1e-3) and --Y_loss_weight specifies the weight of Y-loss loss (default=1e5). You can change the values of these weight and play with them to create different photos. --swapweight specifies the weight of swap loss (default=1e1). You can change the values of these weight and play with them to create better photos.

Acknowledgement

We encode our code build upon LouieYang/deep-photo-styletransfer-tf

Citation

If you find this code useful for your research, please cite:

@misc{YangPhotoStyle2017, author = {Yang Liu}, title = {deep-photo-style-transfer-tf}, publisher = {GitHub}, organization={Alibaba-Zhejiang University Joint Research Institute of Frontier Technologies}, year = {2017}, howpublished = {\url{https://github.com/LouieYang/deep-photo-styletransfer-tf}} }

and

@misc{Adaptive medical image deep color perception algorithm2019, author = {Shiyue Tong}, title = {Adaptive medical image deep color perception algorithm}, publisher = {GitHub}, year = {2019}, howpublished = {\url{https://github.com/Tongshiyue/Adaptive-medical-image-deep-color-perception-algorithm}} }

Contact

Feel free to contact me if you have any question ([email protected])