Table 10.

Results of Transfer Learning. Comparison of segmentation performance of the proposed cGAN and U-Net without and with transfer learning and testing on the AMROA testing dataset. Highest DSCs achieved for each tissue are in bold.

Transfer Learning Results
AMROA Testing
Network	Training	F Bone	T Bone	P Bone	F Cartilage	T Cartilage	P Cartilage	VM Muscle	GM Muscle	ACL	PCL
cGAN	AMROA	0.971 ± 0.005	0.953 ± 0.012	0.947 ± 0.007	0.849 ± 0.046	0.804 ± 0.024	0.869 ± 0.053	0.812 ± 0.066	0.869 ± 0.069	0.618 ± 0.140	0.613 ± 0.143
	SKI10	0.940 ± 0.024	0.947 ± 0.013		0.735 ± 0.005	0.561 ± 0.190
	OAI ZIB	0.962 ± 0.009	0.951 ± 0.010		0.817 ± 0.032	0.790 ± 0.014
	SKI10 → AMROA	0.970 ± 0.008	0.961 ± 0.004	0.940 ± 0.001	0.871 ± 0.029	0.774 ± 0.039	0.858 ± 0.038	0.922 ± 0.037	0.897 ± 0.057	0.586 ± 0.043	0.468 ± 0.186
	OAI ZIB → AMROA	0.972 ± 0.003	0.962 ± 0.001	0.947 ± 0.001	0.875 ± 0.026	0.811 ± 0.042	0.879 ± 0.022	0.908 ± 0.053	0.909 ± 0.077	0.664 ± 0.058	0.652 ± 0.112
	AMROA → SKI10	0.954 ± 0.015	0.949 ± 0.005		0.761 ± 0.025	0.544 ± 0.085
	AMROA → OAI ZIB	0.960 ± 0.007	0.951 ± 0.012		0.821 ± 0.042	0.815 ± 0.015
U-Net	AMROA	0.972 ± 0.006	0.960 ± 0.001	0.941 ± 0.010	0.886 ± 0.007	0.834 ± 0.010	0.890 ± 0.034	0.000 ± 0.000	0.000 ± 0.000	0.643 ± 0.153	0.641 ± 0.008
	SKI10	0.937 ± 0.031	0.944 ± 0.026		0.754 ± 0.009	0.637 ± 0.044
	OAI ZIB	0.959 ± 0.003	0.953 ± 0.010		0.820 ± 0.026	0.798 ± 0.012
	SKI10 → AMROA	0.974 ± 0.003	0.965 ± 0.000	0.947 ± 0.004	0.879 ± 0.012	0.815 ± 0.016	0.896 ± 0.031	0.000 ± 0.000	0.000 ± 0.000	0.665 ± 0.114	0.000 ± 0.000
	OAI ZIB → AMROA	0.973 ± 0.004	0.964 ± 0.005	0.948 ± 0.005	0.893 ± 0.010	0.817 ± 0.043	0.898 ± 0.011	0.000 ± 0.000	0.000 ± 0.000	0.648 ± 0.104	0.000 ± 0.000
	AMROA → SKI10	0.950 ± 0.031	0.959 ± 0.002		0.758 ± 0.010	0.681 ± 0.009
	AMROA → OAI ZIB	0.962 ± 0.006	0.951 ± 0.010		0.813 ± 0.032	0.790 ± 0.039

SKI10/OAI ZIB → AMROA: Pretraining the network for 20 epochs on the SKI10/OAI ZIB dataset followed by network fine-tuning for 80 epochs on the AMROA dataset.

AMROA → SKI10/OAI ZIB: Pretraining the network for 20 epochs on the AMROA dataset followed by network fine-tuning for 80 epochs on the SKI10/OAI ZIB dataset.

Abbreviations: FB – femoral bone, TB – tibial bone, PB – patellar bone, FC – femoral cartilage, TC – tibial cartilage, PC – patellar cartilage, VM Muscle - vastus medialis muscle, GM Muscle – medial head of gastrocnemius medialis muscle, ACL – anterior cruciate ligament, PCL – posterior cruciate ligament, DSC - Sørensen–Dice similarity coefficient.