View on GitHub

Cartilaginous tumor grade prediction from CT radiomics

Model data developed in WEKA for the classification of cartilaginous tumors on CT images

Download this project as a .zip file Download this project as a tar.gz file

Cartilaginous tumors study data

Model data developed in the Waikato Environment for Knowledge Analysis (WEKA) platform for the classification of cartilaginous tumors using radiomics from CT images The “Files” folder contains the trained model (usable in weka with an appropriate dataset) and a text file with the training-testing run information. In this file, the weights used by the final model are available as well as all predictions and predicted probabilities for each case both in the training (through cross-validation) and test sets. Corresponding confusion matrices and accuracy metrics are also presented.

Preparing the data

The model was trained on radiomic data extracted using PyRadiomics. Using bidimensional masks and the settings file included in the repository, an appropriate test set can be obtained from any CT exam of a cartilaginous tumor. The extracted data should then be scaled by using the pickled Python objects also included in the repository. The pickled feature list contains the feature set on which to apply the scaler, while the scaler object contains the scikit-learn MinMaxScaler that can be loaded and fitted to the new data. Finally, the resulting, scaled dataset should be saved either as a CSV or ARFF file to be read by WEKA. Please note WEKA expects the class to be present in the final column of the dataset, if present.

How to load and use the model in WEKA

The WEKA workbench allows for a user-friendly implementation of machine learning algorithms. Its lower barrier to entry was partly behind its choice for the conduction of our analysis. The software is freely available online at this link together with detailed documentation. To load our model in WEKA, one can follow these steps:

Citation

License

Creative Commons Attribution 4.0 International (CC-BY-4.0). See “LICENSE” file for full details.