UNSTRUCTURED
Background: Mask face is a characteristic clinical manifestation of Parkinson's disease (PD), but subjective evaluations from different clinicians often show low consistency owing to lacking accurate detection technology. With the objective of making monitoring easier and more accessible, we developed a markerless 2D video of facial features recognition based artificial intelligence (AI) model to assess facial features of PD patients and aimed to investigate how AI could help neurologists improve PD early diagnostic performance.
Methods: We collected 140 videos of facial expressions of 70 PD patients and 70 healthy controls from three hospitals. We developed and tested the AI model that performs mask face recognition of PD patients based on the acquisition and evaluation of facial features including geometric features and texture features, using a single 2D video camera. The diagnostic performance of AI model was compared with 5 neurologists.
Results: Experimental results show that our AI models can achieve feasible and effective facial feature recognition ability to assist PD diagnosis. The precision and F1 values of PD diagnosis can reach 83% and 86%, using geometric features and texture features, respectively. When these two features are combined, a F1 value of 88% can be reached. Further, the facial features of patients with PD were not affected by the motor and non-motor symptoms of PD.
Conclusions: PD patients commonly exhibit facial features. Video of facial features recognition based AI model can provide a valuable tool to assist PD diagnosis and potential of realizing remote monitoring on patients’ condition especially on the COVID-19 pandemic.
A novel artificial intelligence protocol to investigate potential leads for Parkinson's disease