 |
Background: Hypomimia is a symptom of Parkinson's disease (PD), characterized by a decrease in facial movement and loss of face emotional expressions. This study aims to detect hypomimia in participants with early-stage PD based on facial action units (AUs). Methods: A total of 299 video recordings were included, consisting of 208 PD subjects and 91 healthy control (HC), asked to perform fast syllable repetitions. To distinguish typical facial muscle movement from PD subjects associated with hypomimia, we compute the AUs derivatives. Global features were extracted based on the AUs intensities and their derivatives, and XGBoost was used to classify PD vs. HC. Results: We obtain classification scores up to 73.00% in terms of balanced accuracy (BA) and 78.38% area under the curve (AUC) at video visit level. These results are promising for detecting hypomimia at an early stage of PD, and this work could potentially allow for continuous monitoring of hypomimia outside of hospitals through telemedicine.
