Table 2.

Prediction error rate for test and train set using four best semantic features with primary nodule

A. Including size-based features
Training
FeaturesAccuracy (error), %SensitivitySpecificityE[AUC] μ, σ (95% CI)E[AUC] Gould (9) μ, σ (95% CI)
1Short axis (cm), contour, concavity, texture81.02 (18.98)0.7620.9170.88, 0.08 (0.69–0.98)0.57, 0.04 (0.49–0.65)
2Long axis (cm), border definition, vascular convergence, lymphadenopathy79.6 (20.4)0.7620.950.87, 0.1 (0.53–0.98)0.58, 0.05 (0.48–0.69)
3Short axis (cm), contour, concavity, nodules in nontumor lobes79.02 (20.98)0.7860.9170.86, 0.09 (0.7–0.98)0.58, 0.06 (0.42–0.66)
4Short axis (cm), contour, concavity, spiculation82.42 (17.58)0.7620.9170.86, 0.08 (0.66–0.98)0.58, 0.05 (0.47–0.66)
5Short axis (cm), contour, spiculation, nodules in nontumor lobes80.9 (19.1)0.7620.9170.81, 0.1 (0.62–0.98)0.59, 0.05 (0.5–0.67)
B. No size-based feature
Training
FeaturesAccuracy (error), %SensitivitySpecificityE[AUC] μ, σ (95% CI)E[AUC] Gould (9) μ, σ (95% CI)
1Location, fissure attachment, lobulation, spiculation73.2 (26.8)0.7380.8170.83, 0.09 (0.68–0.98)0.59, 0.049 (0.48–0.66)
2Location, fissure attachment, spiculation, vascular convergence73.6 (26.4)0.7380.8170.82, 0.09 (0.65–0.96)0.578, 0.06 (0.46–0.69)
3Concavity, border definition, spiculation, perinodule fibrosis69.3 (30.7)0.7140.80.81, 0.09 (0.57–0.95)0.586, 0.056 (0.47–0.7)
4Concavity, border definition, spiculation, texture67.3 (32.7)0.7380.7330.8, 0.1 (0.57–0.98)0.567, 0.05 (0.49–0.67)
5Location, pleural attachment, spiculation, vascular convergence71.5 (28.5)0.7140.8170.8, 0.08 (0.7–0.97)0.587, 0.047 (0.51–0.69)