| [8] |
an enhanced segmentation method based on both Neutrosophic sets (NS) and optimized Fast Fuzzy c-mean (F-FCM) algorithm. |
63 thermograms (29 N / 34 AB) |
SVM Classifier |
Accuracy = 92.06% |
| Precision = 87.5% |
| Recall = 96.55% |
| [23] |
Manual |
40 thermograms (26 N / 14 AB) |
SVM, Naïve Bayes and KNN classifier |
Accuracy = 92.5% and Sensitivity = 78.6% with KNN |
| Accuracy = 85% and Sensitivity = 85.7% with SVM |
| Accuracy = 80% and Sensitivity = 85.7% with Naïve Bayes |
| [24] |
Manual |
68 thermograms (26 Ea / 42 Ac) |
DT, KNN, SVM and SVM-RBF |
Accuracy = 95.59%, |
| Sensitivity = 96% and Specificity = 95.35% with SVM-RBF |
| [25] |
Canny edge detection methods followed by gradient operators and Hough transform for boundary detection |
Thermograms of 22 women (11 N / 11AB) |
SVM Classifier |
Accuracy = 90.91%, |
| Sensitivity = 81.82% |
| Specificity = 100% |
| [26] |
Otsu’s threshold to remove background followed by a reconstruction technique. |
306 thermograms (183 N / 123 AB) |
Feed-forward artificial neural network with gradient decent |
Accuracy = 90.48%, |
| Sensitivity = 87.6%, |
| Specificity = 89.73% |
| [27] |
Manual |
600 thermograms (300 N / 300 AB) |
SVM-C |
Accuracy = 93.5%, |
| Sensitivity = 93%, |
| Specificity = 94% |
| [30] |
Not defined |
282 thermograms (147 N / 135 AB) |
CNN using transfer learning |
Accuracy = 94.3% |
| Precision = 94.7% |
| Recall = 93.3% |
| [32] |
Projection profile analysis |
140 patients (98 N / 32 AB) |
Convolutional Neural Networks optimized by Bayes algorithm |
Accuracy = 98.95% |
|
Proposed method
|
U-Net network |
1000 thermograms (500 N / 500 AB) |
Two-class CNN-based deep learning model |
Accuracy = 99.33%, |
| Sensitivity = 100%, |
| Specificity = 98.67% |
