Research Progress of Artificial Intelligence Image Analysis in Systemic Disease-Related Ophthalmopathy

Table 2

Summary of HR diagnosis model based on deep learning method.

Study	Task	Sample size	AI model	Output
Abbas et al. [23]	Clinical staging diagnosis	1400 images	DenseNet	The sensitivity was 90.5%, the specificity was 91.5%, the accuracy was 92.6%, the score was 92%, and the AUC value was 0.915.
Akbar et al. [24]	Detection and classification	The INSPIRE-AVR and VICAVR datasets and a local dataset	Support vector machine and radial basis function	The average accuracies of the first part were 95.10%, 95.64%, and 98.09%, respectively, and the average accuracies of the second part were 95.93% and 97.50%, respectively.
Arsalan et al. [25]	Detection	The DRIVE, CHASE-DB1, and STARE datasets	A dual-residual-stream-based vessel segmentation network	The sensitivity, specificity, AUC value, and accuracy were as follows: DRIVE: 80.22%, 98.1%, 98.2%, and 96.55%, respectively; CHASE-DB1: 82.06%, 98.41%, 98.0%, and 97.26%, respectively; and STARE: 85.26%, 97.91%, 98.83%, and 96.97%, respectively.
Li et al. [15]	Identification	120002 images	The retinal artificial intelligence diagnosis system	The accuracy was 83.7%.