Invasive Across Patient Movement Decoding Model

Through invasive electrocorticographical recordings of more than 50 Parkinson's disease and epilepsy patients, performing different movement types (rotational handle, button press, hand gripping, clench and release) a movement decoding model was trained and is made publicly available through the following GitHub repository: https://github.com/neuromodulation/AcrossPatientDecodingModel. In the original publication, movement decoding was demonstrated without patient-individual training: "Invasive neurophysiology and whole brain connectomics for neural decoding in patients with brain implants" [1]. 

The machine learning model is a Ridge-Regularized Logistic Regression model, that classifies movement, based on z-score normalized and common-averaged re-referenced FFT features in eight different frequency bands. The required feature estimation can be performed through the py_neuromodulation package: https://github.com/neuromodulation/py_neuromodulation. The respective settings and pre-processing parameters, including an exemplary feature estimation pipeline, are provided in the upper GitHub repository.