/OARs_Seg

Exploring Semantic Segmentation of Thoracic Organs based on U-Net

Primary LanguageJupyter Notebook

OARs_Seg

thoracic organs segmentation using Keras

data_read_preprocess.py contains the following functions:

  • CT file reading.
  • CT DICOM file pre-processing, normalization HU.
  • RT Structure file is read and saved as label map.
  • Store all data in NumPy format according to the original directory structure.

u-net_model.ipynb contains the following functions:

  • Use Keras to build a U-net model.
  • Read the training set files in Google Drive
  • Train the network and derive the model.
  • The visualization of training process.

model_evaluation.ipynb contains the following functions:

  • Load the model and predict the validation set images.
  • Output the predicted masks.
  • Output MIoU on validation set.

Model_B is one of the well-performing models trained by U-Net.

predicted_model_B.png shows the prediction masks of Model B on the validation set