JMIR Publications

ABSTRACT

Background:

Neurodegenerative diseases such as Parkinson’s Disease (PD) produce a gradual generalized loss of motor functions including the ability to contract facial muscles during spontaneous and voluntary emotional expressions, and voluntary non-emotional facial movements. This reduced ability leads to a loss of facial expressiveness which generates a signature “mask-like” appearance of the disease also known as hypomimia.

Objective:

We show that modern computer vision techniques can be applied to detect masked facies and quantify medication states in PD.

Methods:

We collected clinical interviews of PD patients in their ON and OFF motor states, as well as journalistic interviews of the actor Alan Alda obtained before and after he was diagnosed with PD. We trained a convolutional neural network on hundreds of thousand facial images extracted from videos of self-identified persons with PD, along with videos of controls, in order to detect PD-specific facial cues. This learned model was applied to (a) ON/OFF clinical interviews, and (b) pre/post-diagnosis Alan Alda interviews

Results:

The accuracy of the video-based model to separately classify ON vs. OFF states in the clinical samples was 63%, in contrast to an accuracy of 46% when using clinical rater scores for facial PD symptoms. Additionally, Alan Alda’s interviews were successfully classified as occurring before versus after his diagnosis with 100% accuracy

Conclusions:

This work demonstrates that automated facial expression analysis may be a promising adjunctive screening tool for PD masked facies and for following a patient’s motor state.