Table 21.

Multi-class (CO (Class 0) vs. PD Stage 2 (Class 2) vs. PD Stage 2.5 (Class 2.5), and PD Stage 3 (Class 3)) and two-class classification (CO vs. PD (Class 2, Class 2.5, and Class 3)) summary using several CNN classifiers (AlexNet, ResNet-50, ResNet-101, and GoogLeNet) with 10-fold cross-validation for Ju dataset.

Dataset	vGRF Signal	CNN Classifier	Time Window	Frequency Range	Classification Task		Evaluation Parameters
							Sen (%)	Spec (%)	Acc (%)	AUC
Ju	LF	ResNet-50	10 s	0.83–1.95 Hz	Multi-Class	Class 0	96.98	99.67	99.20	0.9833
		ResNet-50	10 s	0.83–1.95 Hz		Class 2	96.59	98.44	97.86	0.9751
		AlexNet	10 s	1.95–50 Hz		Class 2.5	97.39	96.52	96.88	0.9695
		AlexNet	30 s	0.83–1.95 Hz		Class 3	97.14	98.65	98.49	0.9790
		ResNet-101	10 s	0.83–1.95 Hz	Two-Class		98.49	98.95	98.84	0.9968
	RF	AlexNet	10 s	1.95–50 Hz	Multi-Class	Class 0	97.99 *	99.24 *	99.02 *	0.9861 *
		ResNet-101	10 s	0.83–1.95 Hz		Class 2	97.73 *	97.40 *	97.50 *	0.9756 *
		ResNet-101	10 s	0.83–1.95 Hz		Class 2.5	95.87	96.36	96.16	0.9612
		ResNet-50	15 s	0.83–1.95 Hz		Class 3	95.83	98.48	98.22	0.9716
		GoogLeNet	10 s	0.83–1.95 Hz	Two-Class		97.97	98.72	98.57	0.9895
	CF	AlexNet	10 s	1.95–50 Hz	Multi-Class	Class 0	98.49	98.91	98.84	0.9870
		ResNet-50	10 s	0.83–1.95 Hz		Class 2	96.02	98.18	97.50	0.9710
		AlexNet	10 s	1.95–50 Hz		Class 2.5	96.52 *	97.88 *	97.32 *	0.9720 *
		AlexNet	10 s	1.95–50 Hz		Class 3	100 *	99.01 *	99.11 *	0.9951 *
		ResNet-101	10 s	0.83–1.95 Hz	Two-Class		98.94 *	99.04 *	99.01 *	0.9993 *

Note: * denotes the best classification result and was selected using Youden’s index criteria.