Table 8.

Quantitative comparison of the proposed attention model with state-of-the-art attention models for semantic segmentation. The experiments are grouped by the dataset. Results are shown for both individual Jun Ma and Mosmed data, also on the set formed by combining these two sources.

S No	Dataset	Method	DSC	IoU	Precision	Sensitivity	Specificity	AUC
1	Jun Ma dataset [30]	FocusNet [34]	75.67	66.38	73.64	77.17	99.67	73.45
		Dual Attention Network [35]	80.15	70.61	77.82	81.49	99.72	79.09
		Asymmetric Non-local networks [36]	81.12	71.78	80.16	82.08	99.73	82.03
		Multi-scale self-guided attention [37]	86.67	75.31	88.42	84.05	99.75	84.45
		Criss Cross Attention [38]	85.58	74.60	82.84	88.12	99.75	83.21
		Semi Inf Net [8]	88.45	76.07	90.47	85.11	99.78	86.55
		Proposed CNN	88.01	75.03	85.57	90.05	99.77	86.74
2	MosMedData [31]	FocusNet [34]	73.49	63.23	71.22	75.88	99.70	71.54
		Dual Attention Network [35]	75.02	61.00	74.82	75.70	99.71	72.10
		Asymmetric Non-local networks [36]	82.17	69.19	83.25	80.67	99.74	81.67
		Multi-scale self-guided attention [37]	80.97	68.78	80.24	81.33	99.72	77.34
		Criss Cross Attention [38]	82.32	70.05	84.68	80.92	99.74	80.64
		Semi Inf Net [8]	83.23	72.55	85.76	79.61	99.74	82.50
		Proposed CNN	83.71	71.51	82.43	84.58	99.75	81.49
3	Combined dataset	FocusNet [34]	73.81	62.13	68.41	80.15	99.71	71.95
		Dual Attention Network [35]	77.39	64.16	74.59	80.42	99.68	76.23
		Asymmetric Non-local networks [36]	81.96	66.08	80.25	83.74	99.72	78.76
		Multi-scale self-guided attention [37]	82.05	71.17	79.47	84.79	99.75	80.49
		Criss Cross Attention [38]	83.85	72.54	79.68	88.47	99.73	82.75
		Semi Inf Net [8]	84.56	72.32	80.50	89.05	99.74	83.71
		Proposed CNN	85.43	73.44	81.23	89.88	99.74	84.57