Data Homogeneity Effect in Deep Learning-Based Prediction of Type 1 Diabetic Retinopathy

Table 1

Summary of the prediction performance of different transfer learning models in predicting diabetic retinopathy.

	Trained on Kaggle training set						Trained on T1D training set
	Tested on T1D testing set			Tested on Kaggle testing set			Tested on T1D testing set			Tested on Kaggle testing set
	AUC	SEN	SPE	AUC	SEN	SPE	AUC	SEN	SPE	AUC	SEN	SPE
DenseNet-121	0.86	0.77	0.79	0.74	0.67	0.71	0.91	0.81	0.86	0.55	0.55	0.54
InceptionV3	0.86	0.74	0.79	0.74	0.62	0.74	0.87	0.73	0.86	0.59	0.56	0.59
VGG16	0.88	0.78	0.82	0.77	0.66	0.75	0.84	0.67	0.84	0.54	0.59	0.49
Xception	0.86	0.74	0.82	0.71	0.60	0.72	0.88	0.74	0.90	0.59	0.61	0.52

T1D: type 1 diabetes; AUC: area under the curve; SEN: sensitivity; SPE: specificity.