Table 1.
Comparison of different state-of-the-art machine learning studies on Parkinson’s disease.
Fig 1.
Experimental framework proposed for the development of this work.
* SVM 1, SVM 2, and SVM 3 classify between PD and Healthy Control (HC).
Table 2.
Demographic and clinical information of the participants included in the FacePark-GITA database.
Fig 2.
Facial expression stages according to the elicited valence measured with the Affectiva tool.
(left) Healthy woman, 63 years old; (right) Woman with Parkinson’s disease, 67 years old, FE item = 2.
Table 3.
Results of classification using a single image from the extracted image sequence.
Table 4.
Results of the classification using different combinations of the extracted frames sequences.
Fig 3.
Action Units defined for the Experiment 2.
Table 5.
FAU detection results of the VGGFace2 model after retraining with the EmotioNet database.
Table 6.
PD classification results using the Freeze 75 model.
Table 7.
PD classification results using the Freeze 50 model.
Fig 4.
PD classification ROC curves obtained from the different input sequences in the retrained Freeze models.
Table 8.
FAU detection results of the VGG-8 and ResNet-7 training with EmotioNet database.
Table 9.
PD classification results using the VGG-8 model.
Table 10.
PD classification results using the ResNet-7 model.
Fig 5.
Comparison between PD classification ROC curves obtained using the NOnAOffN sequence in the Freeze 75, ResNet-7 and VGG-8.
Table 11.
Classification results using a CNN architecture trained from scratch with the NOnAOffN sequences.
Table 12.
PD classification results of classification with the Triplet 75 model.
Table 13.
PD classification results of classification with the Triplet 50 model.
Table 14.
PD classification results using the Triplet-VGG8 model.
Table 15.
PD classification results using the Triplet-ResNet7 model.
Fig 6.
(Up) Principal components spaces generated from the features of the different models and (Bottom) score distributions of PD patients and Healthy Control (HC) subjects obtained by the SVM classifier.
Table 16.
Pairwise comparison results with Mann-Whitney U test and Bonferroni correction for the classification models.
Fig 7.
Heatmaps representations in two different tasks over the three domains studied: (a) Right wink task and (b) Smile task.