Table 3 Performance of the experts and models in classifying high-frequency lesions in PBTs and bone infections in the internal test set
From: Deep learning models in classifying primary bone tumors and bone infections based on radiographs
Bone tumors | n# | EG1 | EG2 | EG3 | Expert average | E3 | E4 | ViT | SWIN | Ensemble | Model average |
|---|---|---|---|---|---|---|---|---|---|---|---|
Osteochondroma | 90 | 95.6% | 93.3% | 93.9% | 94.3% | 100.0% | 100.0% | 100.0% | 100.0% | 100.0% | 100.0% |
Osteosarcoma | 49 | 70.4% | 91.8% | 77.4% | 79.9% | 89.8% | 95.9% | 93.9% | 95.9% | 95.9% | 94.3% |
Fibrous dysplasia | 49 | 85.7% | 89.8% | 92.9% | 89.5% | 77.6% | 81.6% | 83.7% | 95.9% | 87.8% | 85.3% |
GCT | 46 | 94.6% | 95.7% | 91.3% | 93.8% | 91.3% | 91.3% | 91.3% | 89.1% | 87.0% | 90.0% |
Bone infections | N | EG1 | EG2 | EG3 | Expert average | E3 | E4 | ViT | SWIN | Ensemble | Model average |
Bone TB | 170 | 56.5% | 57.4% | 86.9% | 66.9% | 85.3% | 86.5% | 84.1% | 85.3% | 90.6% | 86.4% |
Osteomyelitis | 77 | 50.6% | 53.2% | 62.9% | 54.1% | 50.6% | 48.7% | 46.8% | 57.14% | 66.2% | 54.8% |
