Table 5.
Performance comparison of different CAD systems
| CAD System | Year | Dataset | Extracted features | Classifier | Accuracy (%) | AUC |
|---|---|---|---|---|---|---|
| Santosh et al. [26] | 2016 |
Montgomery dataset (MC), Shenzhen dataset (SZ) |
Thoracic edge map encoding | Neural network |
79.23 (MC) 86.36 (SZ) |
0.88 (MC) 0.93 (SZ) |
| Lopes et al. [28] | 2017 |
MC dataset, SZ dataset |
CNN (GoogleNet, ResNet and VggNet ) transfer learning | SVM |
82.6 (MC) 84.7 (SZ) |
0.926 (MC) 0.926 (SZ) |
| Santosh et al. [12] | 2018 |
MC dataset, SZ dataset, Indian (IN) dataset |
Texture, Shape, Edge, Symmetry |
Bayesian network, Neural network, Random forest (RF) |
83 (MC) 86 (IN) 91 (SZ) |
0.90 (MC) 0.94 (IN) 0.96 (SZ) |
| Vajda et al. [10] | 2018 |
MC dataset, SZ dataset |
Set A: IH, GM, SD, CD, HOG, LBP Set B: Color, intensity, edge, shape, texture Set C: Eccentricity, centroid, bounding box, orientation, extent, size |
Neural network |
78.3 (MC) 95.57 (SZ) |
0.87 (MC) 0.99 (SZ) |
| Rajaraman et al. [41] | 2018 |
SZ MC Kenya (K) India (I) |
Ensemble (pretrained CNNs, HOG, GIST, SURF) | SVM, logistic regression |
0.934 (SZ) 0.875 (MC) 0.776 (K) 0.960 (I) |
0.991 (SZ) 0.962 (MC) 0.826 (K) 0.965 (I) |
| Pasa et al. [32] | 2019 |
MC dataset, SZ dataset, Combined (MC and SZ) dataset, Belarus dataset |
Custom CNN |
79 (MC) 84.4 (SZ) 86.2 (combined) |
0.811 (MC) 0.90 (SZ) 0.925 (combined) |
|
| Govindarajan et al. [27] | 2019 | MC dataset |
Bag of features (BoF) approach with speeded-up Robust feature (SURF) descriptor |
Multilayer perceptron | 87.8 | 0.94 |
| Kyung et al. [42] | 2020 |
Chest X-ray 14 dataset (for training/validation), MC dataset (for testing), SZ dataset (for Testing), Johns Hopkins Hospital dataset (JHH) (for testing) |
ResNet-50 (transfer learning) | CNN |
0.91 (SZ) 0.87 (JHH) |
|
| Sahlol et al. [43] | 2020 |
SZ dataset dataset 2 |
MobileNet, Feature selection by AEO |
CNN |
90.2 (SZ) 94.1 (dataset 2) |
|
| Proposed method | 2020 |
MC dataset, SZ dataset |
Ensemble (pre trained CNN, Gabor filter) | Logistic regression, CNN |
93.47 (MC) 97.59 (SZ) |
0.97 (MC) 0.99 (SZ) |
Abbreviations: Intensity Histogram (IH), Gradient Magnitude Histogram (GM), Shape Descriptor Histogram (SD), Curvature Descriptor Histogram (CD), Histogram of Oriented Gradient (HOG), Local Binary Pattern (LBP), First order statistical feature (FOSF), Gray level co-occurrence matrix (GLCM) features, Artiicial Ecosystem-based Optimization (AEO)