Table 1.

Performance of ML algorithms in discriminating GH7 CBHs and EGs^a

Features	Decision tree			Logistic regression			K-nearest neighbor			Support vector machine
	Sens.	Spec.	Acc.	Sens.	Spec.	Acc.	Sens.	Spec.	Acc.	Sens.	Spec.	Acc.
A1	98.6 ± 1.2	45.9 ± 5.0	72.3 ± 3.2	42.0 ± 16.5	52.8 ± 7.3	46.9 ± 6.4	86.5 ± 15.1	88.7 ± 5.4	87.6 ± 5.6	97.0 ± 1.8	85.5 ± 3.4	91.2 ± 2.0
A2	65.9 ± 43.5	37.4 ± 42.2	50.7 ± 3.7	49.3 ± 46.7	50.3 ± 45.3	47.4 ± 2.8	4.6 ± 2.3	97.0 ± 1.9	50.8 ± 3.8	89.2 ± 27.3	18.4 ± 26.0	53.0 ± 3.5
A3	89.0 ± 26.3	16.9 ± 24.8	52.5 ± 3.7	50.8 ± 47.9	49.4 ± 45.4	47.6 ± 3.2	3.0 ± 2.0	97.8 ± 1.6	50.4 ± 3.7	96.7 ± 11.2	11.4 ± 10.9	53.9 ± 3.4
A4	95.7 ± 2.1	99.5 ± 0.7	97.6 ± 1.1	95.8 ± 2.0	99.5 ± 0.6	97.7 ± 1.1	95.8 ± 2.1	99.7 ± 0.5	97.8 ± 1.1	95.6 ± 2.2	99.6 ± 0.6	97.6 ± 1.1
B1	96.8 ± 1.8	44.1 ± 5.5	70.5 ± 3.3	79.3 ± 35.6	34.5 ± 12.2	55.9 ± 13.4	1.3 ± 1.8	98.7 ± 1.6	50.0 ± 3.7	95.1 ± 2.6	72.3 ± 4.4	83.7 ± 2.6
B2	94.6 ± 2.4	99.1 ± 1.2	96.9 ± 1.3	94.7 ± 2.4	98.4 ± 1.2	96.6 ± 1.3	95.3 ± 2.3	97.4 ± 1.8	96.4 ± 1.3	94.8 ± 2.4	98.4 ± 1.6	96.6 ± 1.4
B3	92.4 ± 2.7	99.8 ± 0.5	96.1 ± 1.4	89.9 ± 3.3	99.8 ± 0.5	94.8 ± 1.7	96.3 ± 2.2	98.6 ± 1.1	97.5 ± 1.2	89.7 ± 3.3	99.8 ± 0.4	94.8 ± 1.7
B4	97.9 ± 1.8	98.2 ± 1.3	98.0 ± 1.0	98.2 ± 1.4	98.2 ± 1.2	98.2 ± 0.9	97.6 ± 2.0	98.3 ± 1.3	98.0 ± 1.1	97.8 ± 1.6	98.2 ± 1.3	98.0 ± 1.0
All 8 loops	98.8 ± 1.2	99.1 ± 1.1	98.9 ± 0.8	98.3 ± 1.4	99.2 ± 0.9	98.8 ± 0.8	97.1 ± 2.0	99.4 ± 0.7	98.2 ± 1.1	99.0 ± 1.1	98.9 ± 1.1	98.9 ± 0.7

^aEach ML model was trained separately with each of the eight loops as a single, independent feature, and then with all eight loops combined (last row). The performance was evaluated by measuring the sensitivity (sens.), specificity (spec.) and accuracy (acc.) in percent. Error represents one standard deviation from the mean.