Research Article
End to End Multitask Joint Learning Model for Osteoporosis Classification in CT Images
Table 2
Comparison with the state-of-the-art baselines on dataset.
| Models | Accuracy | Sensitivity | Specificity | F1-score |
| AlexNet (2012) [43] | Normal | 0.868 | 0.785 | 0.785 | 0.878 | Osteopenia | 0.799 | 0.793 | 0.793 | 0.687 | Osteoporosis | 0.931 | 0.882 | 0.882 | 0.756 |
| VGG-19 (2014) [44] | Normal | 0.856 | 0.994 | 0.765 | 0.847 | Osteopenia | 0.756 | 0.795 | 0.655 | 0.825 | Osteoporosis | 0.900 | 0.894 | 0.941 | 0.939 |
| GoogLeNet (2015) [45] | Normal | 0.899 | 0.976 | 0.849 | 0.886 | Osteopenia | 0.811 | 0.871 | 0.655 | 0.869 | Osteoporosis | 0.911 | 0.902 | 0.980 | 0.947 |
| ResNet-50 (2016) [41] | Normal | 0.911 | 0.874 | 0.936 | 0.888 | Osteopenia | 0.868 | 0.894 | 0.802 | 0.907 | Osteoporosis | 0.956 | 0.995 | 0.986 | 0.976 |
| ResNet-101 (2016) [41] | Normal | 0.938 | 0.958 | 0.924 | 0.925 | Osteopenia | 0.871 | 0.917 | 0.750 | 0.911 | Osteoporosis | 0.933 | 0.940 | 0.882 | 0.961 |
| DenseNet-121 (2017) [46] | Normal | 0.897 | 0.982 | 0.840 | 0.884 | Osteopenia | 0.849 | 0.841 | 0.871 | 0.889 | Osteoporosis | 0.952 | 0.967 | 0.843 | 0.972 |
| ShuffleNet (2018) [47] | Normal | 0.926 | 0.970 | 0.896 | 0.913 | Osteopenia | 0.856 | 0.911 | 0.716 | 0.902 | Osteoporosis | 0.931 | 0.924 | 0.980 | 0.959 |
| EfficientNet (2019) [48] | Normal | 0.926 | 0.976 | 0.892 | 0.913 | Osteopenia | 0.871 | 0.904 | 0.784 | 0.910 | Osteoporosis | 0.945 | 0.943 | 0.961 | 0.968 |
| Joint framework (ours) | Normal | 0.971 | 0.964 | 0.976 | 0.964 | Osteopenia | 0.933 | 0.970 | 0.836 | 0.954 | Osteoporosis | 0.957 | 0.962 | 0.922 | 0.975 |
|
|
The bold value indicates that this is the best model results.
|