Table 3 Comparison with state-of-the-art DL-based methods (with a minimum test set size of 500) for a KOA severity assessment task
Year | Method | DL algorithm | Test set size | Accuracy for KLG 0 (%) | Accuracy for KLG 1 (%) | Accuracy for KLG 2 (%) | Accuracy for KLG 3 (%) | Accuracy for KLG 4 (%) | Average class accuracy (%) |
|---|---|---|---|---|---|---|---|---|---|
2016 | Reference [24] | CNN | 2686 | 71 | 20 | 56 | 76 | 80 | 59.60 |
2017 | Reference [25] | CNN | 4400 | 86.9 | 6.0 | 60.2 | 73.0 | 78.1 | 62.29 |
2018 | Reference [26] | Deep Siamese CNN | 5960 | 78 | 45 | 52 | 70 | 88 | 66.70 |
2019 | Reference [27] | CNN ensemble | 1890 | Not available | 69.50 | ||||
2019 | Reference [28] | CNN | 1495 | Not available | 64.3 | ||||
2019 | Reference [5] | Modified CNN | 1385 | 89.8 | 55.6 | 82.6 | 36.0 | 100.0 | 74.3 |
2019 | Reference [10] | CNN ensemble | 5941 | 83.7 | 70.2 | 68.9 | 86.0 | 78.4 | |
2020 | Reference [29] | CNN | 1175 | 79 | 52 | 58 | 59 | 85 | 66.0 |
2020 | Reference [3] | CNN ensemble | 7599 | 94 | 61 | 90 | 96 | 97 | 87.0 |
2020 | Reference [6] | CNN ensemble | 11,743 | 63.0 | 11.0 | 79.8 | 84.8 | 94.9 | 68.0 |
2020 | Reference [4] | CNN | 4090 | 86.5 | 27.0 | 66.8 | 80.9 | 85.8 | 71.0 |
2022 | Reference [7] | ResNet-50 + AlexNet + TL | 634 | 99.8 | 99.4 | 99.5 | 99.6 | 99.6 | 98.9 |
2024 | Ours | PIM ensemble | 17,040 | 85 | 46 | 69 | 82 | 93 | 75.7 |