Table 3.

Performance metrics from machine learning-based classification of ultrasomic features

Features	Adaboost			Random forest			Support vector machine
	AUC	Sensitivity (%)	Specificity (%)	AUC	Sensitivity (%)	Specificity (%)	AUC	Sensitivity (%)	Specificity (%)
CR, ORF and CEMF	0.85 ± 0.01	87.5	76.9	0.85 ± 0.01	87.5	76.9	0.85 ± 0.01	93.8	69.2
CR and CEMF	0.82 ± 0.04	59.4	100	0.83 ± 0.02	71.9	92.3	0.80 ± 0.03	81.3	76.9
ORF and CEMF	0.84 ± 0.02	92.9	71.4	0.85 ± 0.03	92.9	71.4	0.82 ± 0.04	100	71.4
CR and ORF	0.78 ± 0.03	59.4	100	0.78 ± 0.03	56.3	92.3	0.79 ± 0.04	81.3	84.6
CR	0.68 ± 0.06	43.8	100	0.72 ± 0.05	50.0	92.3	0.71 ± 0.05	90.6	46.2
ORF	0.77 ± 0.02	81.3	69.2	0.73 ± 0.04	62.5	84.6	0.74 ± 0.04	87.5	69.2
CEMF	0.75 ± 0.03	78.1	69.2	0.77 ± 0.04	84.4	76.9	0.74 ± 0.06	90.6	53.9

Note: Performance metrics are validation results from hold-out samples (based on ten-fold cross-validation). Data of AUCs in the table are mean ± standard deviation

CR conventional radiomics, ORF original radiofrequency, CEMF contrast-enhanced micro-flow, AUC area under the receiver-operating characteristic curve