RT Journal Article
SR Electronic
T1 Hypomimia detection with a smartphone camera as a possible self-screening tool for Parkinson disease (P3.047)
JF Neurology
JO Neurology
FD Lippincott Williams & Wilkins
SP P3.047
VO 90
IS 15 Supplement
A1 Seliverstov, Yury
A1 Diagovchenko, Dmitrii
A1 Kravchenko, Michael
A1 Babin, Mikhail
A1 Fedotova, Ekaterina
A1 Belyaev, Mikhail
YR 2018
UL http://n.neurology.org/content/90/15_Supplement/P3.047.abstract
AB Objective: To develop automated algorithm for hypomimia detection in patients with Parkinson disease (PD) using video records obtained with smartphones’ cameras.Background: By 2018, over a third of the world’s population is projected to own a smartphone which is almost 2.53 billion smartphone users in the world. These devices which are used in our everyday life allow to capture various kinds of data including video records. PD is the second most common neurodegenerative disorder. It affects 1–2 per 1000 of the population at any time. Hypomimia is one of the cardinal clinical features of PD often presented in its early stages. Early diagnosis is critical for better treatment outcomes. However, there is still a lack of validated, easy-to-use, and relatively cheap assessment methods for hypomimia. Implementation of self-screening techniques may improve timely diagnosis of hypomimia associated conditions such as PD, depression, etc.Design/Methods: PD patients (n=20; Hoehn & Yahr stages 2–2.5) and healthy controls (n=41) matched for age and gender underwent a set of facial expression tests for lower facial muscles; of those, 14 PD patients and 11 healthy controls underwent additional tests for upper facial muscles. Facial movements were recorded using either a front or back camera of either iOS or Android based smartphones. All videos were then converted to 720p 24fps format. We used a gradient boosting classifier based on Scikit-learn software machine learning library. Our algorithm automatically detected particular facial action points and analyzed their kinematics. Leave-one-out cross-validation was implemented.Results: We obtained the following results: precision = 0.93; recall = 0.87; accuracy = 0.88 (classifiaction algorithm returned 1 false negative and 2 false positive subjects).Conclusions: The results of our pilot study demonstrate relatively high quality control values allowing us to consider further development of our hypomimia detection approach. Suggested approach can be considered as a promising self-screening tool for hypomimia-associated conditions.Disclosure: Dr. Seliverstov has nothing to disclose. Dr. Diagovchenko has nothing to disclose. Dr. Kravchenko has nothing to disclose. Dr. Babin has nothing to disclose. Dr. Fedotova has nothing to disclose. Dr. Belyaev has nothing to disclose.