Table 6 Performance of the deep learning model for the type of a meniscal tear
Model | Acc | Pre | Rec | Sen | Spe | AUC (95% CI) | Time (sec) | |
|---|---|---|---|---|---|---|---|---|
Horizontal tear | MobileNet | 52.09% | 41.48% | 70% | 70% | 41.48% | 0.542 (0.463–0.621) | 2.45 |
Ours | 72.23% | 59.3% | 63.75% | 63.75% | 74.07% | 0.761 (0.694–0.828) | 1.26 | |
Complex tear | MobileNet | 64.07% | 32.05% | 78.12% | 78.12% | 60.74% | 0.759 (0.682–0.835) | 1.91 |
Ours | 91.02% | 81.48% | 68.75% | 68.75% | 96.3% | 0.850 (0.759–0.941) | 1.01 | |
Radial tear | MobileNet | 63.09% | 15.25% | 64.29% | 64.29% | 62.96% | 0.651 (0.517–0.785) | 1.76 |
Ours | 72.48% | 15.38% | 42.86% | 42.86% | 75.56% | 0.601 (0.433–0.768) | 0.95 | |
Longitudinal tear | MobileNet | 66.24% | 21.82% | 54.55% | 54.55% | 68.15% | 0.680 (0.561–0.798) | 1.71 |
Ours | 81.53% | 40.54% | 68.18% | 68.18% | 83.7% | 0.858 (0.787–0.930) | 1.03 |