Table 4 Comparison of the classification results of the true microcalcifications as BI-RADS benign or suspicious using temporal subtraction (TS) of mammograms and using only the most recent mammograms (RM), in a leave-one-patient-out cross-validation scheme
Classifier | Sensitivity (%) | Specificity (%) | Accuracy (%) | AUC | |||||||
|---|---|---|---|---|---|---|---|---|---|---|---|
9-Nearest neighbors | TS RM | 96/114 73/114 | (84.2) (64.0) | TS RM | 393/515 357/515 | (76.3) (69.3) | TS RM | 489/629 430/629 | (77.7) (68.4) | TS RM | 0.80 0.67 |
Decision trees | TS RM | 66/114 65/114 | (57.9) (57.0) | TS RM | 446/515 445/515 | (86.6) (86.4) | TS RM | 512/629 510/629 | (81.4) (81.1) | TS RM | 0.72 0.72 |
Random forest | TS RM | 73/114 70/114 | (64.0) (61.4) | TS RM | 461/515 452/515 | (89.5) (87.8) | TS RM | 534/629 522/629 | (84.9) (83.0) | TS RM | 0.77 0.75 |
Multilayer perceptron | TS RM | 93/114 69/114 | (81.6) (60.5) | TS RM | 411/515 302/515 | (79.8) (58.6) | TS RM | 504/629 371/629 | (80.1) (59.0) | TS RM | 0.81 0.6 |
Adaptive boosting | TS RM | 86/114 80/114 | (75.4) (70.2) | TS RM | 433/515 430/515 | (84.1) (83.5) | TS RM | 519/629 510/629 | (82.5) (81.1) | TS RM | 0.8 0.77 |
Bagging | TS RM | 69/114 65/114 | (60.5) (57.0) | TS RM | 458/515 450/515 | (88.9) (87.4) | TS RM | 527/629 515/629 | (83.8) (81.9) | TS RM | 0.75 0.72 |
Gradient boosting | TS RM | 82/114 77/114 | (71.9) (67.5) | TS RM | 447/515 438/515 | (86.8) (85.1) | TS RM | 529/629 515/629 | (84.1) (81.9) | TS RM | 0.79 0.76 |
Ensemble voting | TS RM | 93/114 90/114 | (81.6) (79.0) | TS RM | 475/515 430/515 | (92.2) (83.5) | TS RM | 568/629 520/629 | (90.3) (82.7) | TS RM | 0.87 0.81 |
Neural network | TS RM | 89/114 83/114 | (78.1) (72.8) | TS RM | 450/515 485/515 | (87.4) (94.2) | TS RM | 539/629 568/629 | (85.7) (90.3) | TS RM | 0.83 0.83 |