Study Feature/input Classifier/method Database Performance Giger et al. [123 ] Geometric features Comparison of geometric features Thick-slice diagnostic CT scans of 8 patients with 47 nodules Sensitivity of 94% with1.25 FPs per case Kanazawa et al. [81 ] 8 features Rule based scheme Helical CT scans from 450 patients with 230 nodules Sensitivity of 90% Armato et al. [9 , 124 ] Nine 2D and 3D features Rule-based scheme and LDA Thick-slice (10 mm) diagnostic CT scans of 43 patients with 171 nodules Sensitivity of 70% with 42.2 FPs per case in a leave-one-out test Lee et al. [71 ] 13 features Rule-based scheme and LDA Thick-slice (10 mm) diagnostic CT scans of 20 patients with 98 nodules Sensitivity of 72% with 30.6 FPs per case Ko and Betke [64 ] Location and 2 shape features (circularity and roundness) Rule-based scheme Helical CT scans of 16 studies (8 initial and 8 followup) obtained from 8 patients with 370 nodules Sensitivity of 86% Brown et al. [87 ] Prior models based on 4 features Fuzzy matching Thick slice (5–10 mm) CT scans of 17 patients with 36 nodules Sensitivity of 86% and 11 FPs per case Wiemker et al. [72 ] 4 shape and intensity features NA Thin-slice (1 mm) HRCT scans of 50 subjects (a total of more than 20,000 image sections); Sensitivity of 86% with 4.9 FPs per case for nodules with diameter ≥1 mm and sensitivity of 95% with 4.4 FPs per case with diameters ≥2 mm Gurcan et al. [78 ] Six 2D and 3D features Rule-based scheme and LDA Thick-slice (2.5–5 mm, mostly 5 mm) diagnostic CT scans of 34 patients with 63 nodules Sensitivity of 84% with 74.4 FPs per case in a leave-one-out test Suzuki et al. [111 ] Pixel values in a
subregion Multiple MTANNs Thick-slice (10 mm) screening LDCT scans of 63 patients with 71 nodules with solid, partially solid, and nonsolid patterns, including 66 cancers Sensitivity of 80.3% with 4.8 FPs per case in a validation test Mekada et al. [63 ] Minimum directional difference filter Rule-based scheme CT scans of 6 subjects with 361 nodules (160–350 sections per case) Sensitivity of 71% and 7.4 FPs per case Arimura et al. [116 ] Pixel values in a
subregion for MTANNs (selected features for LDA) Rule-based scheme followed by multiple MTANNs (or LDA with Wilks' lambda stepwise feature selection) 106 thick-slice (10 mm) screening LDCT scans of 73 patients with 109 cancers with solid, partially solid, and nonsolid patterns Sensitivity of 83% with 5.8 FPs per case in a validation test (or a leave-one-out test for LDA) Awai et al. [74 ] 6 geometric features Artificial neural network classier CT scans of 82 patients with 78 nodules (a total of 3,556 image sections) Sensitivity of 80% with 0.87 FPs per section Paik et al. [69 ] SNO method that describes the shape and geometry Rule-based scheme CT scans of 8 patients Sensitivity of 90% with 5.6 FPs per case in a cross validation test Farag et al. [125 , 126 ] NA Template modeling approach using LS Thin-slice (2.5 mm) screening LDCT scans of 16 patients with 119 nodules and 34 normal patients Sensitivity of 93% with 3.4 FPs per case Ge et al. [127 ] 44 features including 3D gradient field descriptors and ellipsoid features LDA with Wilks' lambda stepwise feature selection 82 thin-slice (1.0–2.5 mm) CT scans of 56 patients with 116 solid nodules Sensitivity of 80% with 14.7 FPs per case in a leave-one-out test Mendonca et al. [70 ] Geometric and intensity models combined with eigen curvature analysis Rule-based scheme Thin-slice (1.25 and 2.5 mm) CT scans of 242 exams from two institutions: 50 CT scans with 109 nodule and 192 CT scans with 210 nodules Sensitivity of 67.5% and 9.3 FPs per case for data from the first 50 CT scans and sensitivity of 62.9% and 10.3 FPs per case for the second 192 CT scans in a leave-one-out test Matsumoto et al. [128 ] 8 features LDA Thick-slice (5 or 7 mm) diagnostic CT scans of 5 patients (4 of which used contrast media) with 50 nodules Sensitivity of 90% with 64.1 FPs per case in a leave-one-out test Yuan et al. [129 ] NA ImageChecker CT LN-1000 by R2 Technology Thin-slice (1.25 mm) CT scans of 150 patients with 628 nodules Sensitivity of 73% with 3.2 FPs per case in an independent test Pu et al. [130 ] NA Scoring method based on the similarity distance combined with a marching cube algorithm Thin-slice (2.5 mm) screening CT scans of 52 patients with 184 nodules including 16 nonsolid nodules Sensitivity of 81.5% with 6.5 FPs per case Retico et al. [131 ] Pixel values in a subvolume Voxel-based neural approach (MTANN) Thin-slice (1 mm) screening CT scans of 39 patients with 102 nodules Sensitivities of 80–85% with 10–13 FPs per case Ye et al. [15 ] 15 features Rule-based scheme followed by a weighted SVM Thin-slice (1 mm) screening CT scans of 54 patients with 118 nodules including 17 non-solid nodules Sensitivity of 90.2% with 8.2 FPs per case in an independent test Golosio et al. [132 ] 42 features from multithreshold ROI Fixed-topology ANN Thin-slice (1.5–3.0 mm) CT scans of 83 patients with 148 nodules that one radiologist detected from the LIDC database Sensitivity of 79% with 4 FPs per case in an independent test Murphy et al. [133 ] Features selected from 135 features KNN Thin-slice screening CT scans of 813 patients with 1,525 nodules Sensitivity of 80% with 4.2 FPs per case in an independent test Messay et al. [134 ] Features selected from 245 features LDA and quadratic discriminant analysis with feature selection Thin-slice CT scans of 84 patients with 143 nodules from the LIDC database Sensitivity of 83% with 3 FPs per case in a 7-fold cross-validation test Tan et al. [135 ] 45 features Feature-selective classifier based on a genetic algorithm and ANNs Thin-slice CT scans of 125 patients with 80 nodules that 4 radiologists agreed from the LIDC database Sensitivity of 87.5% with 4 FPs per case in an independent test Riccardi et al. [136 ] Maximum intensity projection data from the volume of interest Heuristic approach (rule-based scheme) Thin-slice CT scans of 154 patients with 117 nodules that Sensitivity of 71% with 6.5 FPs per case in a 2-fold cross-validation test
