Neurodegenerative diseases are characterized by a spectrum of disruptive
symptoms affecting daily life. These symptoms include tremor - involuntary, rhythmic alternating spasms of agonist and antagonist muscle groups. The MDS-UPDRS scale is used for the assessment of motor and non-motor symptoms of Parkinson’s Disease, including tremor. Spirography provides qualitative and quantitative assessment of tremor in diseases such as Parkinson’s disease, Essential tremor and dystonia. ParkinsonCheck, a mobile application, aids tremor detection and classification (Parkinsonian vs. Essential). This research’s focal point is developing predictive models for predicting the MDS-UPDRS scores for postural, kinetic, and resting tremor, by using spirographic data gathered with ParkinsonCheck. Feature extraction methods that we used are identical to the ones used for developing the ParkinsonCheck models. For modeling we explored Random Forest, Logistic
Regression, and XGBoost, feature subset selection involved ReliefF and mutual information. Encouraging outcomes were observed for binary classification of postural tremor with ROC AUC of 0.754 (Random forest) and kinetic tremor with ROC AUC of 0.713 (XGBoost trained on a subset selected with mutual information). Resting tremor, due to its nature and inability to be captured by the ParkinsonCheck application was not predictable.
|