Table 5.
Diagnostic performance of detecting bone metastases in bone scans of previous studies.
| Authors | Type of CNN | Primary tumors | n | Prv (%) | Acc (%) | Sen (%) | Spe (%) | Prc (%) | Rec (%) |
|---|---|---|---|---|---|---|---|---|---|
| Papandrianos et al. [13] | 4-layer CNN | Prostate | 778 | 41.9 | 91.6 | 92.7 | 96.0 | ||
| Papandrianos et al. [14] | 3-layer CNN | Breast | 408 | 54.1 | 92.5 | 94.0 | 92.0 | 93.4 | 93.8 |
| Pi et al. [15] | Inception-V3 | Lung (31 %) Breast (24 %) Prostate (10 %) Other (12 %) Benign (22 %) |
15,474 | 37.5 | 95.0 | 93.2 | 96.1 | ||
| Hsieh et al. [16] | ResNet50V2 | Breast (59 %) H&N (12 %) Prostate (7 %) Lung (5 %) Liver (3 %) Other (14 %) |
37,427 | 8.1 | 96.1 | 59.9 | 99.3 | 87.8 | |
| Guo et al. [17] | 26-layer CNN | Lung | 945 | 64.8 | 83.1 | 87.0 | 87.0 | ||
| Han et al. [18] | GLUE 2D-CNN | Prostate | 9133 | 32.7 | 90.0 | 82.8 | 93.5 | ||
| Liu et al. [19] | ResNet 34 | Prostate Lung Breast Gastrointestinal |
621 | 43.9 | 88.6 | 92.6 | 85.5 |
Acc, accuracy; CNN, convolutional neural network; GLUE, global–local unified emphasis; H&N, head and neck; Prc, precision, Prv, prevalence; Rec, recall; Sen, sensitivity; Spe, specificity.