Challenges with the diagnosis and monitoring of Parkinson’s disease and dementia with Lewy bodies (DLB) are a major hindrance for accurate, early diagnosis and treatment planning. Reduced facial expression is a classical clinical feature of PD and may also be present in patients with dementia with Lewy bodies. Early facial behavior changes are often overlooked or misinterpreted. The often subtle onset of facial behavior changes combined with the lack of objective methods for assessment hinders early identification and monitoring.
In this study, we will develop a deep learning system to objectively compare facial behavior in people with dementia with Lewy bodies, Parkinson’s disease, and age-matched control participants. Participants will undergo in-person facial behavior data collection through a web-based tool. Participants will undergo eight facial expression tests (one posed facial expression test and seven spontaneous facial expression tests) while their faces are recorded by a webcam on a laptop through the web-based tool. We will then train deep learning algorithms that take the facial videos as input and automatically predict disease state. We will analyze the performance of the deep learning algorithms using metrics that include accuracy, sensitivity, and specificity. We will also analyze which facial behavior features are most important for predicting people with Parkinson’s disease and dementia with Lewy bodies.
Our study will determine if facial behavior can be used in the future as a digital biomarker for people with dementia with Lewy bodies and Parkinson’s disease, which could have profound potential in telemedicine visits, remote monitoring of disease severity, and as a tool for differential diagnosis. We hope to pave the way toward the development of a low-cost, scalable decision support tool to aid clinicians in differential diagnosis and disease monitoring for Parkinson’s disease and dementia with Lewy bodies.