Deep learning-based thigh muscle segmentation for reproducible fat fraction quantification using fat–water decomposition MRI

Insights into Imaging

Table 1 Segmentation accuracy assessed by Dice coefficients between manual and automated segmentation for each ROI of each thigh volume in the testing set

	ROI1	ROI2	ROI3	ROI4
Dataset 1 (mean ± SD as repeated scans were performed)
1st, left	0.942 ± 0.002	0.886 ± 0.007	0.866 ± 0.015	0.915 ± 0.004
1st, right	0.923 ± 0.010	0.812 ± 0.038	0.866 ± 0.011	0.900 ± 0.014
2nd, left	0.927 ± 0.006	0.848 ± 0.005	0.846 ± 0.024	0.917 ± 0.004
2nd, right	0.927 ± 0.013	0.872 ± 0.007	0.877 ± 0.013	0.902 ± 0.024
3rd, left	0.933 ± 0.003	0.869 ± 0.007	0.861 ± 0.003	0.925 ± 0.003
3rd, right	0.932 ± 0.011	0.874 ± 0.004	0.876 ± 0.021	0.915 ± 0.014
Dataset 2
1st, left	0.958	0.890	0.881	0.905
1st, right	0.954	0.918	0.897	0.911
2nd, left	0.923	0.848	0.819	0.842
2nd, right	0.920	0.774	0.835	0.824
3rd, left	0.944	0.806	0.819	0.904
3rd, right	0.946	0.814	0.771	0.871
4th, left	0.960	0.914	0.931	0.897
4th, right	0.959	0.904	0.902	0.884
Mean	0.939	0.859	0.860	0.894
SD	0.015	0.044	0.040	0.029

ROI1, quadriceps femoris; ROI2, sartorius; ROI3, gracilis; ROI4, hamstrings; SD, standard deviation