Table 4 Comparing the state-of-the-art retinal diseases recognition methods based on ViT architecture. It should be emphasized that all presented percentages values are averages.

Model	Number of classes	Dataset	Dataset size	Accuracy (%)	Precision (%)	Recall (%)	F1-score (%)	Specificity (%)	Study
ViT	4	OCT⁴⁶	35,432	95.14	95.24	97.72	96.42	n/d	¹¹
T2T-ViT				96.01	96.15	98.34	97.23	n/d
Mobile ViT				99.17	99.59	99.59	99.58	n/d
MedViT	3	NEH-v2²⁵	16,822	99.70	n/d	n/d	n/d	93.05	²⁵
MedViT	4	UCSD³⁴	108,312	84.10	n/d	n/d	n/d	98.65	²⁵
ViT	4	OCT2017	84,484	97.50	97.90	n/d	97.30	96.02	⁴⁷
SViT	4	OCT2017	84,484	99.90	100	n/d	99.95	97.29	⁴⁷
MedViT	4	RetinaMNIST	1,600	97.47	n/d	n/d	n/d	n/d	⁴⁸
HCTNet	4	OCT2017	84,484	91.56	99.60	98.02	98.60	96.55	⁴⁹
Conv-ViT	4	OCT2017	84,484	92.37	94.00	94.00	94.00	94.00	³¹
ViT	4	OCT2017	84,484	99.06					⁵⁰
SwinT	4	OCT2017	84,484	98.01	99.07	99.07	99.07	n/d	⁵¹
LLCT				98.70	97.83	97.65	99.23	n/d
SwinPolyT				99.82	99.80	99.80	99.80	n/d
0.3cmFD-CNN	3	Duke	3231	97.19	96.37	97.34	96.75	98.73	¹⁹
0.3cmFD-CNN	4	UCSD	84,484	99.40	99.40	99.41	99.40	99.41	¹⁹
D-KNN	3	Duke	3231	96.88	95.90	97.10	96.38	98.62	¹⁹
D-KNN	4	UCSD	84,484	99.50	99.51	99.50	99.50	99.83	¹⁹
D-SVM	3	Duke	3231	97.50	96.61	97.64	97.03	98.91	¹⁹
D-SVM	4	UCSD	84,484	99.60	99.60	99.60	99.60	99.87	¹⁹
FT-CNN	3	Duke	3231	99.69	99.76	99.70	99.73	99.90	²⁰
FT-CNN	4	UCSD	84,484	99.70	99.70	99.90	99.70	99.90	²⁰
R-FTCNN	3	Duke	3231	99.06	98.81	98.91	98.86	99.55	¹⁵
R-FTCNN	4	UCSD	85,484	99.60	99.60	99.60	99.60	99.87	¹⁵
RS-A ViT	3	RS-A ViT	1280	96.92	96.06	96.02	96.02	98.32	this study

Quick links

Search