|
| Related work | Classifier method | Dataset | Features | Accuracy of normal/abnormal classifier | Accuracy of benign/malignant classifier | CAD tool based-GUI |
|
| Kamil Jassam [39] | KNN | MIAS | Gray level cooccurrence matrix (GLCM) | None | 86.1% | No |
| Al-Azzam and Shatnawi [40] | Compare supervised learning (SL) with semisupervised learning (SSL) for 9 algorithms | Wisconsin diagnostic breast cancer (WDBC) dataset | 30 attributes of 569 patients with 569 instances | None | K-NN (SL = 98% & SSL = 97%) and logistics regression (SL = 97% & SSL = 98%) | None |
| Huang et al. [41] | SVM with three functions and two features: bagging and boosting | Two datasets | First dataset: 11 attributes with 699 instances
Second dataset: 117 attributes with 102294 instances | None | 96.85%, 95% | None |
| Cai et al. [45] | Conventional neural network (CNN) | The datasets were collected at two medical institutions | 990 images, 540 malignant masses, and 450 benign lesions | None | 87.68% | No |
| Zebari et al. [47] | Five classifiers with ANN | Mini-MIAS, and other datasets | (i) LBP
(ii) FD
(iii) Propose M-FD | None | Train, test 79.11, 83.61 80.06, 89.55 96.2, 100 | No |
| Kamil and Jassam [39] | SVM | MIAS | GLCM | None | 83.5% | |
| Our work | Two-level SVM classifiers | MIAS dataset | Statistical classes, wavelet class, and SFS technique: 21 attributes with 307 instances | 100% | 87.1% | Yes |
|
