Table 3 Comparison of the classification results of the possible microcalcifications as normal tissue or radiologically true microcalcifications with temporal subtraction (TS) of mammograms or 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 | 501/629 469/629 | (79.7) (74.6) | TS RM | 4928/5739 4593/5739 | (85.9) (80.0) | TS RM | 5429/6368 5062/6368 | (85.3) (79.5) | TS RM | 0.83 0.76 |
Decision trees | TS RM | 458/629 392/629 | (72.8) (62.3) | TS RM | 5311/5739 5234/5739 | (92.5) (91.2) | TS RM | 5769/6368 5626/6368 | (90.6) (88.4) | TS RM | 0.83 0.78 |
Random forest | TS RM | 484/629 421/629 | (77.0) (66.9) | TS RM | 5565/5739 5591/5739 | (97.0) (97.4) | TS RM | 6049/6368 6012/6368 | (95.0) (94.4) | TS RM | 0.87 0.82 |
Multilayer perceptron | TS RM | 510/629 470/629 | (81.1) (74.7) | TS RM | 4806/5739 4275/5739 | (83.7) (74.5) | TS RM | 5316/6368 4745/6368 | (83.5) (74.5) | TS RM | 0.82 0.73 |
Adaptive boosting | TS RM | 510/629 473/629 | (81.1) (75.2) | TS RM | 5561/5739 4659/5739 | (88.0) (81.2) | TS RM | 6071/6368 5132/6368 | (87.3) (80.6) | TS RM | 0.85 0.77 |
Bagging | TS RM | 473/629 408/629 | (75.2) (64.9) | TS RM | 5510/5739 5528/5739 | (96.0) (96.3) | TS RM | 5983/6368 5936/6368 | (94.0) (93.2) | TS RM | 0.86 0.80 |
Gradient boosting | TS RM | 512/629 468/629 | (81.4) (74.4) | TS RM | 5291/5739 5073/5739 | (92.2) (88.4) | TS RM | 5803/6368 5541/6368 | (91.1) (87.0) | TS RM | 0.87 0.80 |
Ensemble voting | TS RM | 512/629 457/629 | (81.4) (72.8) | TS RM | 5480/5739 5536/5739 | (95.5) (96.5) | TS RM | 5992/6368 5993/6368 | (94.1) (94.1) | TS RM | 0.88 0.83 |
Neural network | TS RM | 518/629 384/629 | (82.4) (61.1) | TS RM | 4911/5739 5058/5739 | (85.6) (88.1) | TS RM | 5429/6368 5448/6368 | (85.3) (85.5) | TS RM | 0.84 0.76 |