Multitask Interactive Attention Learning Model Based on Hand Images for Assisting Chinese Medicine in Predicting Myocardial Infarction

Table 5

Evaluation result of different attention module combinations.

Models	Component				Task	Result (%)
	1	2	3	4		Accuracy	Sensitivity	Specificity	AUC
One-AM-1	√				M	51.24	53.45	49.36	52.39
					P	50.75	52.61	50.13	52.11
One-AM-2		√			M	51.78	54.14	51.01	54.21
					P	53.16	56.73	51.84	55.59
One-AM-3			√		M	57.28	59.31	53.82	58.37
					P	54.53	59.46	52.44	58.84
One-AM-4				√	M	62.21	64.57	58.30	64.73
					P	62.55	63.79	60.15	65.03
Two-AM-1	√	√			M	56.41	59.73	53.51	57.32
					P	54.89	56.56	51.74	56.56
Two-AM-2	√		√		M	62.81	65.82	58.66	64.39
					P	62.17	66.15	60.59	62.95
Two-AM-3	√			√	M	66.15	69.81	62.54	68.26
					P	65.23	67.48	62.05	68.47
Two-AM-4		√	√		M	62.79	65.22	59.57	65.01
					P	61.97	63.98	57.18	64.56
Two-AM-5		√		√	M	67.09	70.02	65.91	71.63
					P	65.77	67.57	61.10	69.05
Two-AM-6			√	√	M	73.80	74.46	67.09	75.79
					P	70.19	73.11	64.96	74.14
Three-AM-1	√	√	√		M	76.21	78.40	69.38	79.52
					P	74.26	77.03	69.65	77.31
Three-AM-2	√	√		√	M	73.46	77.73	71.09	76.37
					P	73.10	74.95	70.33	73.96
Three-AM-3	√		√	√	M	72.59	74.30	68.72	75.93
					P	73.33	76.24	70.65	77.01
Three-AM-4		√	√	√	M	76.68	80.33	75.03	79.36
					P	78.21	82.06	74.29	80.65
Four-AM (ours)	√	√	√	√	M	81.92	83.47	79.93	82.30
					P	82.46	84.11	80.33	83.27