Table 1 Performance of the classifier A (shallow neural network in the case of 3D RI tomograms and logistic regression in the case of 2D QPMs) and the classifier B (shallow neural network in the case of 3D RI tomograms and linear discriminant in the case of 2D QPMs) over the test set.
Metric | Classifier A | Classifier B | ||||||
|---|---|---|---|---|---|---|---|---|
3D RI tomogram | 2D QPM | 3D RI tomogram | 2D QPM | |||||
MC | Tumor cell | NB | OC | NB | OC | MC | Tumor cell | |
Accuracy | 97.4 | 87.3 | 97.3 | 85.2 | ||||
True positive rate (sensitivity or recall) | 95.7 | 97.8 | 96.7 | 98.3 | 82.8 | 90.0 | 63.8 | 93.4 |
True negative rate (specificity) | 97.8 | 95.7 | 98.3 | 96.7 | 90.0 | 82.8 | 93.4 | 63.8 |
Positive predictive value (precision) | 91.8 | 98.9 | 99.2 | 93.7 | 94.4 | 72.0 | 71.4 | 90.9 |
Negative predictive value | 98.9 | 91.8 | 93.7 | 99.2 | 72.0 | 94.4 | 90.9 | 71.4 |
Balanced accuracy | 96.8 | 78.6 | 97.5 | 86.4 | ||||
F1 score | 93.8 | 98.3 | 97.9 | 95.9 | 88.2 | 86.4 | 67.4 | 92.1 |
Matthews correlation coefficient | 92.1 | 59.7 | 93.9 | 69.5 | ||||
Fowlkes–Mallows index | 93.8 | 98.3 | 97.9 | 96.0 | 88.4 | 80.5 | 67.5 | 92.1 |
