Table 1.
Feature-based algorithms applied to the Lung Image Database Consortium Image Collection (LIDC-IDRI) database.
| Author | Year | CT Scans Incl. | Accuracy (%) | Sensitivity (%) | Specificity (%) | AUC | Classifier | Nodule Type | Selected Features |
|---|---|---|---|---|---|---|---|---|---|
| Akram et al. * [7] | 2015 | 84 | 96.6 | 96.9 | 96.3 | 0.980 | SVM | All types | 2D and 3D geometric and intensity statistical features |
| Alilou et al. * [8] | 2014 | 60 | NA | 80.0 | NA | NA | SVM | Solid | 2D and 3D subset of features |
| Bai et al. [9] | 2015 | 99 | NA | 80.0 | NA | NA | NA | All types | Local shape analysis and data-driven local contextual feature learning |
| Choi et al. * [10] | 2014 | 84 | 99.0 | 97.5 | 97.5 | 0.998 | SVM-r | All types | CAD system for different dimensions of AHSN features |
| El Regaily et al. [11] | 2017 | 400 | 70.5 | 77.7 | 69.5 | NA | The simple rule classifier | All types | Geometric and intensity statistical features |
| Firmino et al. * [12] | 2016 | 420 | NA | 94.4 | NA | NA | SVM | All types | HOG; watershed; features of texture, shape, and appearance |
| Gonçalves et al. * [13] | 2018 | NA | 68.4 | 55.0 | 87.5 | 0.905 | SVM | Solid nodules | Intensity-, texture-, and shape-based features |
| Gong et al. * [14] | 2016 | 100 | 91.5 | 90.2 | 91.5 | 0.960 | FLDA | Not GGO | 11 selected image features |
| Gupta et al. [15] | 2017 | 899 | NA | 90.0 | NA | 0.980 | softmax | Large nodules | Feature mapping: stacked sparse autoencoder (SSAE) |
| Hancock et al. * [16] | 2017 | 619 | 88.0 | 84.6 | NA | 0.949 | Nonlinear | All types | Nonlinear classifier, diameter, and volume features included |
| Jaffar et al. [17] | 2018 | 59 | 98.8 | 98.4 | 98.7 | 0.999 | Random forest | All types | Novel ensemble shape gradient features (NESGF) |
| Liu et al. [18] | 2017 | 107 | NA | 89.4 | NA | NA | NA | All types | Geometric and statistical features |
| Lu et al. [19] | 2015 | 98 | NA | 85.2 | NA | NA | Regression tree | All types | Hybrid scheme based on 16 features |
| Naqi et al. * [20] | 2018 | 250 | 99.0 | 98.6 | 98.2 | 0.990 | SVM | All types | Geometric texture features descriptor (GTFD) |
| Shaukat et al. * [21] | 2017 | 850 | 97.1 | 98.1 | 96.0 | 0.995 | SVM-Gaussian | All types | Intensity, shape (2D and 3D), and texture features |
| Taşcı et al.* [22] | 2015 | 24 | 92.9 | NA | NA | 0.883 | GLMR | Juxtapleural | Seven shape- and texture-based features |
| Wang et al. * [23] | 2018 | NA | 95.9 | 95.6 | 95.0 | 0.961 | SS-ELM | All types | Haralick features and morphological features |
| Zhang et al. * [24] | 2018 | 71 | NA | 89.3 | NA | NA | SVM | Juxtavascular nodules | 3D skeletonization |
| Zhao et al. [25] | 2017 | NA | 91.2 | NA | NA | 0.970 | softmax | All types | Global and local features |
CAD: Computer-aided detection, AHSN: angular histograms of surface normal, HOG: Histogram of oriented Gradients, NA: not available. The studies marked with a star (“*”) presented several types of alterations to the algorithm, producing different results. These results are not presented in the table.