Classification of Parkinsonian Rigidity Using AdaBoost with Decision Stumps

Parkinson's disease (PD) is the second most frequent neurodegenerative disease. The clinical manifestations of PD mainly contain tremor, bradykinesia, rigidity, and loss of postural reflexes, in which rigidity responds immediately upon PD treatment. Currently, the Parkinsonian rigidity assessment depends mostly upon subjective judgment of neurologists in accordance with their experience, which shows low consistency among different evaluators. Currently, few works have been done for Parkinson's rigidity estimation, and the existing work does not achieve strong evaluation performance. In this paper, we designed an electromechanical driving device to obtain the parameters correlated well with the rigidity symptom. Using these parameters as inputs, we employed the AdaBoost algorithm to score the rigidity severity objectively. In the multiclass classification model, decision stumps were used as weak classifiers to provide decision rules for classification. By combining handling of noise, the AdaBoost model shows great robustness (classification accuracy of IO-fold cross-validation: 99.609%). Compared with KNN and LIBSVM methods, the proposed model could achieve superior classification accuracy (classification accuracy: 97.097%), better stability (at least 57.807% higher than KNN), and more reasonable computational time (2.72 times faster than LIBSVM).