This code supports the findings in the manuscript titled 'Machine learning for (non-)epileptic tissue detection from the intraoperative electrocorticogram' by Hoogteijling et al. Please cite this article for every work that uses code or data from this repository.
The code was developed in Pyhton 3.9.2 using Spyder 4.2.1. The following packages were used and can be installed via pip install or conda install:
- Pandas (1.5.1)
- Numpy (1.21.5)
- sklearn (1.3.0)
- matplotlib.pyplot (3.3.4)
- shap (0.40.0)
Please read the manuscript for data aquisition and pre-processing details. The intraoperative electrocorticogram spectral bands power data set can be download from DataVerse (see manuscript).
Training and test set data are organized in a .CSV file where one column represent the label and every other column represent a spectral feature. Each row represents a single ioECoG 20-second epoch, similar to:
| Index | Label | Spectral feature 1 | Spectral feature 2 | ... |
|---|---|---|---|---|
| 0 | Label | Feature value | Feature value | ... |
| 1 | Label | Feature value | Feature value | ... |
| 2 | Label | Feature value | Feature value | ... |
| ⋮ | ⋮ | ⋮ | ⋮ | ⋱ |
For the training .CSV file, the last column represents the fold the 20-second epoch was allocated to for five-fold cross-validation.
Make a copy of the loadData_example.py and name it loadData.py. In loadData.py, specify the path to the folder containing the Xy_train and Xy_test .CSV files at line 14.
Run mainSH.py for ETC performance in five-fold cross-validation and on the test set. The last cell in this code will show the SHAP analysis.