Table 5 Comparison of computation time for different MA detection methods
Methods | Highlights | Average time | Databases used |
|---|---|---|---|
Derwin [26] | Texture descriptors, SVM | 29 s | One database, in resolutions of \(768\times 576\), \(1058\times 1061\) and \(1389\times 1383\) |
Chudzik [31] | FCN | 220 s | e-ophtha MA and DIARETDB1, with FROC score of 0.562 and 0.369 |
Dashtbozorg [22] | Local convergence index features, RUSBoosting | 3 min | e-ophtha MA and DIARETDB1, with FROC score of 0.546 and 0.547 |
Wang [24] | Singular spectrum analysis, KNN | 1 min | DIARETDB1 database, with Sensitivity of 0.517 at 1 FPI |
Habib [38] | Tree ensemble | 65 s | DIARETDB1 database, with FROC score of 0.2109 |
Seoud [17] | Dynamic shape features, RF | 98 s for range of 2000–3000 pixels | DIARETDB1 database, with Sensitivity of 0.6 at 6 FPI |
Proposed method | DLC feature, NB | 29 s for \(2544\times 1696\), 3 s for \(1400\times 960\), 2.6 s for \(1500\times 1152\) | e-ophtha MA and DIARETDB1, with FROC score of 0.374 and 0.210 |