Towards Identification of Hypomimia in Parkinson’s Disease Based on Face Recognition Methods

Towards Identification of Hypomimia in Parkinson’s Disease Based on Face Recognition Methods

Paper in proceedings (conference paper)

Hypomimia manifested as an expressionless face with little or no sense of animation is a typical symptom of Parkinson’s disease (PD). Although some researchers tried to quantify and diagnose the hypomimia based on the analysis of video-recordings, a study dealing with a possibility of its identification using the simple static face analysis is missing. The goal of this work is therefore to verify whether PD hypomimia can be detected even from static face images. For this purpose we enrolled 50 PD patients and 50 age- and gender-matched healthy controls. Parameterization based on face recognition methods in combination with conventional classifiers (random forests, XGBoost, etc.) were used to automatically identify PD hypomimia. Among the classifiers, the decision tree algorithm achieved the best accuracy (67.33 %). The results suggest that automatic static face analysis can support PD hypomimia diagnosis, nevertheless is not accurate enough to outperform the approaches based on video-recordings processing.

Hypomimia manifested as an expressionless face with little or no sense of animation is a typical symptom of Parkinson’s disease (PD). Although some researchers tried to quantify and diagnose the hypomimia based on the analysis of video-recordings, a study dealing with a possibility of its identification using the simple static face analysis is missing. The goal of this work is therefore to verify whether PD hypomimia can be detected even from static face images. For this purpose we enrolled 50 PD patients and 50 age- and gender-matched healthy controls. Parameterization based on face recognition methods in combination with conventional classifiers (random forests, XGBoost, etc.) were used to automatically identify PD hypomimia. Among the classifiers, the decision tree algorithm achieved the best accuracy (67.33 %). The results suggest that automatic static face analysis can support PD hypomimia diagnosis, nevertheless is not accurate enough to outperform the approaches based on video-recordings processing.

Parkinson’s disease; hypomimia; face recognition; machine learning

Parkinson’s disease; hypomimia; face recognition; machine learning

RAJNOHA, M.; BURGET, R.; MEKYSKA, J.; ELIÁŠOVÁ, I.; KOŠŤÁLOVÁ, M.; REKTOROVÁ, I.

2019

06.11.2018

Moskva

978-1-5386-9361-2

2018 10th International Congress on Ultra Modern Telecommunications and Control Systems and Workshops (ICUMT)

182

185

4

https://ieeexplore.ieee.org/document/8631249

http://hdl.handle.net/

1570493577