Table 3.

Facial-recognition-based diagnosis system for neurodegenerative diseases.

Study	Disease	Data	Sample Size	Method	Efficacy
Bandini et al. [60]	PD	Video	17 PD, 17 HC	Intraface tracking algorithm, Euclidean distance, SVM	Difference (p < 0.05) between PD and HC
Rajnoha et al. [61]	PD	Image	50 PD, 50 HC	Random Forests, XGBoost	Accuracy = 67.33%
Jin et al. [23]	PD	Video	33 PD, 31 HC	Face++ [62], tremor extraction, LSTM neural network	Precision = 86%
Ali et al. [5]	PD	Video	61 PD, 543 HC	OpenFace 2.0 [24], SVM	Accuracy = 95.6%
Hou et al. [63]	PD	Video	70 PD, 70 HC	HOG, LBP, SVM, k-NN, Random Forests	F1 = 88%
Nam et al. [25]	AD	Video	17 AD, 17 HC	OpenFace 2.0 [24], extract movement coordinates to calculate Spearman’s correlation coefficient	Difference (p < 0.05) between AD and HC
Umeda et al. [64]	AD	Image	121 AD, 117 HC	Xception, SENet50, ResNet50, VGG16, and simple CNN with SGD and Adam optimizer	Xception with Adam showed the best accuracy = 94%
Bandini et al. [18]	ALS	Video	11 ALS, 11 HC	AAM, CLM, ERT, SDM, FAN	Accuracy = 88.9%

Abbreviations and explanations: PD, Parkinson’s disease; AD, Alzheimer’s disease; ALS, amyotrophic lateral sclerosis; HC, healthy control; SVM, Support Vector Machines; LSTM, Long Short-Term Memory; HOG, Histogram of Oriented Gradient; LBP, Local Binary Pattern; k-NN, k-Nearest Neighbors; AAM, active appearance models; CLM, constrained local model; ERT, ensemble of regression trees; SDM, supervised descent method; FAN, face alignment network.