Table 3.

Identification and classification of suspicious FDG-avid foci by fully automated neural network when compared with the consensus reader reference standard

	Per lesion	Per patient
Accuracy	95.6 (93.9–97.2) %	97.3 (95.7–98.5) %
Sensitivity	92.1 (89.3–94.8) %	92.3 (79.3–96.6) %
Specificity	97.7 (95.0–98.9) %	97.6 (94.2–99.1) %
Positive predictive value	96.1 (91.4–98.2) %	96.5 (92.1–98.6) %
Negative predictive value	95.3 (90.8–97.4) %	93.2 (86.2–97.1) %

Only PERCIST measurable lesions were regarded for this table