Table 9.

Accuracy comparison of the proposed RIFM model with either geometric or non-geometric-based methods

	RIMONE								SLO Images
Features (i.e. geometric or	TP	FN	TN	FP	Sn	Sp	Acc	p value	TP	FN	TN	FP	Sn	Sp	Acc	p value
non-geometric)
RIFM	36	3	81	4	92.3 %	95.3 %	94.4 %	–	17	2	31	1	89.5 %	96.9 %	94.1 %	–
Geometric based Methods
Geo-metric (Vertical CDR)	29	10	80	5	74.4 %	94.1 %	87.9 %	<0.10	16	3	29	3	84.2 %	90.6 %	88.2 %	=0.28
Geo-metric (Horizontal CDR)	26	13	76	9	66.7 %	89.4 %	82.3 %	<0.01	14	5	28	4	73.7 %	87.5 %	82.4 %	<0.10
Geo-metric (Vasculature Shift)	26	13	75	10	66.7 %	88.2 %	81.5 %	<0.01	14	5	20	12	73.7 %	62.5 %	66.7 %	<0.001
Non-geometric based Methods
Global Features (Mix)	35	4	74	11	89.7 %	87.1 %	87.9 %	<0.10	13	6	28	4	68.4 %	87.5 %	80.4 %	<0.10
Textural Features (Variable Offset) [18, 19]	30	9	71	14	76.9 %	83.5 %	81.5 %	<0.01	11	8	18	12	57.9 %	56.2 %	56.9 %	<0.001
Textural Features (Variable Scale) [18, 19]	35	4	74	11	89.7 %	87.1 %	87.9 %	<0.10	12	7	21	11	63.2 %	65.6 %	64.7 %	<0.005
Textural Features (Scale + Offset) [18, 19]	35	4	74	11	89.7 %	87.1 %	87.9 %	<0.10	13	6	28	4	68.4 %	87.5 %	80.4 %	<0.10
Higher Order Spectra Features [19]	34	5	74	11	87.2 %	87.1 %	87.1 %	<0.05	12	7	24	8	63.2 %	75.0 %	70.6 %	<0.01
Gabor Features [51]	34	5	75	10	87.2 %	88.2 %	87.9 %	<0.10	11	8	24	8	57.9 %	75.0 %	68.6 %	<0.01
Wavelet Features [15]	31	8	65	20	79.5 %	76.5 %	77.4 %	<0.001	11	8	24	8	57.9 %	75.0 %	68.6 %	<0.01
Gaussian Features	32	7	67	18	82.1 %	78.8 %	79.8 %	<0.01	10	9	26	6	52.6 %	81.3 %	70.6 %	<0.05
Dyadic Gaussian Features	28	11	75	10	71.8 %	88.2 %	83.1 %	<0.05	10	9	26	6	52.6 %	81.3 %	70.6 %	<0.05