https://github.com/andreped/t-loss-tf
Robust T-Loss for Medical Image Segmentation with TensorFlow backend
https://github.com/andreped/t-loss-tf
image-analysis keras medical python semantic-segmentation t-loss tensorflow tf tloss
Last synced: 8 months ago
JSON representation
Robust T-Loss for Medical Image Segmentation with TensorFlow backend
- Host: GitHub
- URL: https://github.com/andreped/t-loss-tf
- Owner: andreped
- License: mit
- Created: 2023-08-14T08:15:08.000Z (about 2 years ago)
- Default Branch: main
- Last Pushed: 2023-09-10T16:19:03.000Z (about 2 years ago)
- Last Synced: 2025-02-17T17:52:24.334Z (8 months ago)
- Topics: image-analysis, keras, medical, python, semantic-segmentation, t-loss, tensorflow, tf, tloss
- Language: Python
- Homepage:
- Size: 19.9 MB
- Stars: 3
- Watchers: 2
- Forks: 0
- Open Issues: 0
-
Metadata Files:
- Readme: README.md
- License: LICENSE
Awesome Lists containing this project
README
t-loss-tf
Robust T-Loss for Medical Image Segmentation with TensorFlow backend
[](https://github.com/andreped/t-loss-tf/releases)
[](https://opensource.org/licenses/MIT)
[](https://github.com/andreped/t-loss-tf/actions)
This T-loss implementation is an adaption of the original PyTorch code [Digital-Dermatology/t-loss-loss](https://github.com/Digital-Dermatology/t-loss).
More information about T-loss and the original paper can be found [here](https://robust-tloss.github.io/).
## [Installation](https://github.com/andreped/t-loss-tf#installation)
```
pip install git+https://github.com/andreped/t-loss-tf.git
```
## [Usage](https://github.com/andreped/t-loss-tf#usage)
As the t-loss contains a trainable parameter, in keras the loss needed to be implemented as a custom layer.
Hence, instead of setting the loss as normally through `model.compile(loss=[...])`, just add it to the model
at an appropriate place (e.g., at the end of the network).
An example can be seen below:
```python
import tensorflow as tf
from t_loss import TLoss
# create dummy inputs and GTs
input_shape = (16, 16, 1)
x = tf.ones((32,) + input_shape, dtype="float32")
y = tf.ones((32,) + input_shape, dtype="float32")
# define network
input_x = tf.keras.Input(shape=input_shape)
input_y = tf.keras.Input(shape=input_shape)
z = tf.keras.layers.Conv2D(filters=4, kernel_size=(1, 1), activation="relu")(input_x)
z = tf.keras.layers.MaxPooling2D(pool_size=(2, 2))(z)
z = tf.keras.layers.UpSampling2D(size=(2, 2))(z)
z = tf.keras.layers.Conv2D(filters=1, kernel_size=(1, 1), activation="sigmoid")(z)
z = TLoss(tensor_shape=input_shape, image_size=input_shape[0])(z, input_y)
model = tf.keras.Model(inputs=[input_x, input_y], outputs=[z])
# train model
model.compile(optimizer="adam")
model.fit(x=[x, y], y=y, batch_size=2, epochs=1, verbose="auto")
```
## [License](https://github.com/andreped/t-loss-tf#license)
The code in this repository is released under [MIT License](https://github.com/andreped/t-loss-tf/blob/main/LICENSE).
## [Citation](https://github.com/andreped/t-loss-tf#citation)
The implementation is based on the original torch implementation hosted [here](https://github.com/Digital-Dermatology/t-loss).
Hence, if this code is used, please cite the original research article:
```
@inproceedings{gonzalezjimenezRobustTLoss2023,
title = {Robust T-Loss for Medical Image Segmentation},
author = {Gonzalez-Jimenez, Alvaro and Lionetti, Simone and Gottfrois, Philippe and Gröger, Fabian and Pouly, Marc and Navarini, Alexander},
journal = {Medical {{Image Computing}} and {{Computer Assisted Intervention}} – {{MICCAI}} 2023},
publisher = {{Springer International Publishing}},
year = {2023},
}
```