Quantifying hypomimia in Parkinson patients using a depth camera

Abstract

One of Parkinson’s disease early symptoms is called hypomimia (masked facies), and timely detection of this symptom could potentially assist early diagnosis. In this study we developed methods to automatically detect and assess the severity of hypomimia, using machine learning tools and a 3D sensor that allows for fairly accurate facial movements tracking. To evaluate our prediction of hypomimia score for participants not included in the training set, we computed the score’s correlation with hypomimia scores provided by 2 neurologists. The correlations in 4 conditions were 0.84, 0.69, 0.71, 0.70. This should be compared with the correlation between the somewhat subjectives scores of the two neurologists, which is 0.78. When training classifiers to discriminate between people who suffer from hypomimia and people who do not, the area under the curve of the corresponding Receiver Operating Characteristic curves in the same 4 conditions is 0.90 − 0.99. These encouraging results provide proof of concept that automatic evaluation of hypomimia can be sufficiently reliable to be useful for clinical early detection of Parkinson-related hypomimia.