Machine Learning-Based Technique for the Severity Classification of Sublingual Varices according to Traditional Chinese Medicine

Table 2

Diagnostic performance summary of two ML techniques in conjunction with two feature extraction techniques by using original images as features, compared with the diagnoses by two TCM doctors. The number in square brackets is the 95% CI. For each case, training time is in seconds including the time spent by a feature extraction technique when applied. The numbers in parentheses are speedup for the counterpart method against the one trained with original features.

	Sensitivity (%)	Specificity (%)	Accuracy (%)	PPV (%)	NPV (%)	Time (s)
TCM doctors	94.5 [90.3, 98.7]	85.5 [80.6, 90.4]	90.0 [87.3, 92.7]	87.6 [83.7, 91.5]	94.9 [91.2, 98.5]	N/A
SVM-original	80.5 [74.5, 86.5]	88.5 [84.4, 92.6]	84.5 [81.6, 87.4)	88.3 [84.7, 91.9]	83.1 [78.8, 87.5]	3.8e1 (1.0)
Ridge-original	79.5 [74.1, 84.9]	90.0 [86.6, 93.4]	84.7 [82.3, 87.2]	89.7 [86.4, 93.0]	82.3 [78.8, 85.8]	1.1e0 (1.0)
SVM-PCA + SIR	79.0 [72.1, 85.9]	88.0 [84.1, 91.9]	83.5 [79.8, 87.2]	87.2 [83.5, 90.9]	82.1 [77.2, 87.0]	(2100.0)
Ridge-PCA + SIR	81.0 [75.1, 86.9]	88.0 [84.7, 91.3]	84.5 [81.5, 87.5]	87.5 [84.4, 90.6]	83.3 [79.0, 87.7]	(275.0)
SVM-CNN	91.0 [85.5, 96.5]	81.5 [76.4, 86.6]	86.2 [82.8, 89.7]	83.8 [80.0, 87.6]	91.2 [86.5, 95.8]	(378.0)
Ridge-CNN	90.5 [84.5, 96.5]	84.5 [78.7, 90.3]	87.5 [83.8, 91.2]	86.5 [82.1, 90.9]	91.3 [86.4, 96.2]	(41.0)