Examples#

The following examples illustrate the intended use of PyRaDiSe:

Example Data#

The data from the “Segmentation of Vestibular Schwannoma from Magnetic Resonance Imaging” dataset was used for the given examples. Because this dataset is large with 242 patients in total, a subset of data from 5 patients is made available in a separate GitHub repository such that the reader can execute the given examples in his/her setup. The data is available as DICOM files and as NIfTI files.

The example data contains for each patient a T1-weighted Gd-enhanced MR image, a T2-weighted MR image, and a DICOM-RTSS with segmentations of the current tumor volume, the left cochlea, the skull, and an older version of the tumor volume. The original data includes in addition a second RT Structure Set, two RT Dose, and two RT Plans for each patient that were removed to lower the size of the example data. All images were acquired with a 32-channel 1.5T Siemens Avanto scanner between the years 2012 and 2018. The RT Structure Sets were created manually in consensus between the neurosurgeon and physician using both the T1- and T2-weighted images. The segmentation were performed using the Leksell GammaPlan software from Elekta, Sweden.

In addition to the example patient data, a PyTorch-based DL-model for skull segmentation is provided that is used in the inference example. The model was trained on 50 patients from the mentioned dataset for 15 epochs with a fixed learning rate of 0.0001 and a batch size of 4 images. The Binary Cross Entropy loss function was used for training. Because this model is used for demonstration purposes only and is not intended for clinical use, the model is not trained until convergence.