Automated classifiers for early detection and diagnosis of retinopathy in diabetic eyes

BMC Bioinformatics

Table 5 Model testing results obtained after changing the size of the training data set and using the TR as an input feature

Size of the training data set	20 healthy eyes		30 healthy eyes		40 healthy eyes
TR vs. FD	GCL + IPL (eye/scans)	OPL (eye/scans)	GCL + IPL (eye/scans)	OPL (eye/scans)	GCL + IPL (eye/scans)	OPL (eye/scans)
TP	49/294	48/288	39/234	39/234	29/174	29/174
FN	5/30	6/36	5/30	5/30	5/30	5/30
TN	35/210	37/222	35/210	36/216	35/210	37/222
FP	8/48	6/36	8/48	7/42	8/48	6/36
PPV	0.86	0.89	0.83	0.85	0.78	0.83
Sensitivity	0.91	0.89	0.89	0.89	0.85	0.85
Specificity	0.81	0.86	0.81	0.84	0.81	0.86

Results of sensitivity, specificity, accuracy, predictive values and positive predictive values obtained for the GCL + IPL complex and OPL when training the Bayesian radial basis function network with 20, 30 and 40 healthy eyes with the total reflectance (TR) and fractal dimension (FD) as the input and target features, respectively.

ISSN: 1471-2105