Table 3.
Performance comparison regarding classification accuracy, sensitivity, and specificity (%) with recently published state-of-the-art algorithms.
| Reference | Model | Accuracy | Sensitivity | Specificity | |
|---|---|---|---|---|---|
| Andrea et al. [20] | KNN (1) | 74.05% | - | - | |
| Zhang et al. [21] | CNN (2) | 90.00% | 81.00% | 92.00% | |
| Byra et al. [22] | CNN (3) | 96.30% | 100.00% | 88.20% | |
| Cao et al. [23] | CNN (4) | 73.97% | - | - | |
| Anca et al. [24] | CNN (5) | 93.23% | 88.90% | - | |
| Zamanian et al. [25] | CNN (6) | 98.64% | 97.20% | 100.00% | |
| Proposed methods | Cascaded NN | ♠ | 99.91% | 99.78% | 100.00% |
| ♥ | 100.00% | 100.00% | 100.00% | ||
| ♣ | 99.62% | 99.13% | 100.00% | ||
| ♦ | 100.00% | 100.00% | 100.00% | ||
♠: When being trained and tested by SMC database. ♥: When being trained and tested by Byra database. ♣: When being trained both by SMC and Byra database, but tested by SMC database. ♦: When being trained both by SMC and Byra database, but tested by Byra database. (1) (2012) ANN where k-nearest neighbor is better than SVM. (2) (2019) Shallow convolutional neural network-based model to extract texture feature. (3) (2018) Pretrained CNN through transfer learning. (4) (2019) 3 image-processing techniques: including envelope signal, grayscale values and a NN. (5) (2020) Transfer learning with comparison of 2 pretrained networks: VGG16 and inception V3. (6) (2021) Performance comparison study of 4 pretrained networks: Inception v2, GoogleNet, etc.