Skip to main content
. Author manuscript; available in PMC: 2024 Oct 2.
Published in final edited form as: IEEE Trans Med Imaging. 2023 Oct 2;42(10):2832–2841. doi: 10.1109/TMI.2023.3266137

TABLE II.

COMPARISON WITH STATE-OF-THE-ART METHODS ON KVASIR, ENDOVIS’17, ART-NET AND ROBOTOOL.

Dataset Method DSC IOU MAE F-measure
Kvasir-instrument UNet (fully) 0.901 0.756 0.027 0.862
UNet ++ (fully) 0.893 0.778 0.023 0.875
TransUNet (fully) 0.905 0.855 0.015 0.922
Label ratio la 5% 20% 50% 5% 20% 50% 5% 20% 50% 5% 20% 50%
UNet 0.706 0.730 0.799 0.435 0.508 0.606 0.075 0.055 0.043 0.609 0.674 0.755
UNet++ 0.567 0.736 0.823 0.440 0.612 0.720 0.085 0.041 0.028 0.440 0.760 0.837
TransUNet 0.541 0.753 0.867 0.252 0.706 0.845 0.093 0.029 0.015 0.402 0.826 0.916
Mean Teacher 0.605 0.788 0.892 0.415 0.689 0.799 0.065 0.031 0.020 0.587 0.816 0.888
Deep Co-training 0.489 0.764 0.866 0.292 0.632 0.735 0.084 0.045 0.027 0.452 0.759 0.840
Cross Pseudo 0.709 0.824 0.894 0.607 0.643 0.804 0.051 0.037 0.020 0.755 0.783 0.891
Duo-SegNet 0.403 0.834 0.861 0.274 0.701 0.755 0.081 0.033 0.026 0.430 0.824 0.860
Min-Max Similarity (ours) 0.776 0.874 0.925 0.650 0.768 0.873 0.043 0.024 0.013 0.787 0.868 0.932
EndoVis’ 17 UNet (fully) 0.894 0.840 0.027 0.912
UNet ++ (fully) 0.909 0.841 0.026 0.914
TransUNet (fully) 0.904 0.826 0.029 0.905
Label ratio la 5% 20% 50% 5% 20% 50% 5% 20% 50% 5% 20% 50%
UNet 0.823 0.869 0.885 0.653 0.772 0.819 0.057 0.040 0.029 0.784 0.872 0.902
UNet++ 0.825 0.882 0.890 0.651 0.743 0.760 0.058 0.044 0.041 0.788 0.853 0.864
TransUNet 0.837 0.873 0.882 0.713 0.775 0.790 0.047 0.039 0.035 0.833 0.875 0.882
Mean Teacher 0.875 0.901 0.910 0.797 0.848 0.849 0.037 0.028 0.024 0.885 0.915 0.920
Deep Co-training 0.848 0.895 0.895 0.777 0.845 0.847 0.038 0.026 0.026 0.875 0.913 0.917
Cross Pseudo 0.886 0.909 0.913 0.813 0.850 0.855 0.029 0.025 0.021 0.895 0.919 0.926
Duo-SegNet 0.879 0.906 0.912 0.806 0.849 0.864 0.033 0.025 0.023 0.893 0.918 0.927
Min-Max Similarity (ours) 0.909 0.931 0.940 0.861 0.890 0.899 0.023 0.018 0.017 0.925 0.942 0.947
ART-NET UNet (fully) 0.894 0.752 0.029 0.859
UNet ++ (fully) 0.908 0.799 0.023 0.888
TransUNet (fully) 0.904 0.823 0.019 0.903
Label ratio la 5% 20% 50% 5% 20% 50% 5% 20% 50% 5% 20% 50%
UNet 0.660 0.713 0.812 0.521 0.510 0.679 0.072 0.062 0.038 0.685 0.676 0.809
UNet++ 0.717 0.761 0.866 0.590 0.600 0.743 0.053 0.051 0.030 0.742 0.750 0.852
TransUNet 0.685 0.764 0.841 0.628 0.633 0.733 0.047 0.043 0.032 0.773 0.775 0.846
Mean Teacher 0.747 0.835 0.889 0.614 0.726 0.814 0.051 0.033 0.021 0.761 0.841 0.897
Deep Co-training 0.726 0.820 0.875 0.629 0.714 0.811 0.049 0.033 0.021 0.772 0.833 0.895
Cross Pseudo 0.759 0.824 0.874 0.629 0.708 0.797 0.047 0.035 0.023 0.772 0.829 0.887
Duo-SegNet 0.738 0.771 0.833 0.608 0.664 0.729 0.048 0.039 0.032 0.756 0.798 0.843
Min-Max Similarity (ours) 0.784 0.869 0.917 0.652 0.758 0.843 0.045 0.029 0.017 0.790 0.863 0.915
RoboTool UNet (fully) 0.786 0.617 0.088 0.763
UNet ++ (fully) 0.807 0.656 0.068 0.792
TransUNet (fully) 0.808 0.672 0.063 0.804
Label ratio la 5% 20% 50% 5% 20% 50% 5% 20% 50% 5% 20% 50%
UNet 0.516 0.661 0.730 0.413 0.505 0.558 0.133 0.105 0.093 0.584 0.671 0.716
UNet++ 0.500 0.691 0.734 0.397 0.527 0.568 0.152 0.098 0.087 0.561 0.690 0.724
TransUNet 0.516 0.718 0.732 0.497 0.549 0.559 0.123 0.087 0.090 0.664 0.709 0.717
Mean Teacher 0.575 0.742 0.784 0.443 0.637 0.679 0.137 0.074 0.061 0.614 0.773 0.809
Deep Co-training 0.519 0.714 0.752 0.397 0.593 0.636 0.143 0.080 0.068 0.568 0.744 0.777
Cross Pseudo 0.559 0.711 0.758 0.429 0.593 0.641 0.147 0.083 0.069 0.601 0.745 0.781
Duo-SegNet 0.586 0.701 0.746 0.488 0.556 0.647 0.117 0.086 0.070 0.656 0.715 0.786
Min-Max Similarity (ours) 0.646 0.781 0.831 0.544 0.697 0.750 0.104 0.058 0.046 0.705 0.821 0.857