Table 3.

Predictive performance of ULM compared with clinically available methods.

Method	Individual measurements			Changes between measurements			Combined measurements
	1	2	3	1 → 2	2 → 3	1 → 3	1 & 2	1 & 2 & 3
Accuracy (%)
ULM	75	75	65	85	85	65	80	85
B-mode size measurement	60	60	60	60	60	60	60	60
B-mode image	60	60	60	65	75	60	60	75
Contrast-mode MIP	65	60	65	75	75	65	75	75
Histological CD31 masks	75
AUC
ULM	0.73	0.72	0.47	0.85	0.80	0.69	0.85	0.83
B-mode size measurement	0.31	0.35	0.53	0.21	0.38	0.32	0.34	0.29
B-mode image	0.00	0.26	0.13	0.13	0.75	0.21	0.25	0.76
Contrast-mode MIP	0.57	0.13	0.44	0.79	0.63	0.54	0.77	0.73
Histological CD31 masks	0.83
Sensitivity (%)
ULM	50	63	25	75	88	38	75	75
B-mode size measurement	0	0	0	0	0	0	0	0
B-mode image	0	0	0	13	75	0	0	75
Contrast-mode MIP	38	13	25	50	63	38	50	50
Histological CD31 masks	38
Specificity (%)
ULM	92	83	92	92	83	83	83	92
B-mode size measurement	100	100	100	100	100	100	100	100
B-mode image	100	100	100	100	75	100	100	75
Contrast-mode MIP	83	92	92	92	83	83	92	92
Histological CD31 masks	100
p-value of DeLong's test—AUC of ULM compared to
B-mode size measurement	0.05	0.02	0.77	<0.01	<0.01	0.10	0.01	<0.01
B-mode image	<0.01	0.02	0.05	<0.01	0.90	<0.01	<0.01	0.81
Contrast-mode MIP	0.25	<0.01	0.90	0.47	0.31	0.33	0.37	0.33
Histological CD31 masksa	0.55	0.59	0.04	0.93	0.88	0.28	0.93	1.00

Predictive performance was quantified via accuracy, AUC, sensitivity, and specificity of each method's final model. Additionally, the AUC of the final ULM model was compared with those of the comparative methods using DeLong's test.

^aULM-derived models were compared with the histological CD31 masks obtained before the start of NAC.