Smile and speech for screening Parkinson’s disease and cognitive decline: A crowdsourced cross-sectional study

Abstract Smiling and speech are typically disturbed in patients with Parkinson’s disease (PD). This study aimed to find a subset of smile and speech features suitable for identifying PD and cognitive decline to build predictive models. This crowdsourced cross-sectional study collected smile data, conversational speech, and cognitive tests using a mobile chatbot app from 74 PD patients and 236 healthy controls (HC). In the classification model, the area under the curve for the accuracy of PD classification was 0.944 ± 0.029. In the regression model to predict the cognitive score of PD, linear regression showed a mean absolute error of 1.977 ± 0.309. A machine-learning model using a combination of smile and speech features obtained by a mobile chatbot app could differentiate between PD and HC. Furthermore, a model using a combination of smile and speech features might predict cognitive decline in PD.