Boxplots of the Dice similarity coefficient (DSC) for automated segmentation as compared to the manual delineation. The inputs are A: three modalities, B: T1-weighted MR image only, C: ADC only, and D: FA only.

Contexts in source publication

Context 1

... parameters and characteristics are listed in Table 1. A qualitative comparison of the manually delineated thigh muscles with the structures generated by the deep-learning model is shown in Fig. 1, Panels A, B. Quantitatively, the accuracy of the model was assessed with the Dice similarity coefficient (DSC) and presented with the boxplots in Fig. ...

Context 2

... in the network training on the segmentation accuracy, we trained the model with each of these images separately as the input channels. The best accuracy was achieved with all three modalities included in the training. The most contributing input was from the T1-eighted MRI with the ADC and FA almost equally contributing, see the four panels in Fig. 2. The voxel-wise diffusion tensors and the muscle masks were used with the Hamiltonian Fast-Marching solver to obtain the solution of Eq. (A.4) in the form of isosurfaces of the shortest distances when traveling from a point within the muscle. (Adapted from (27)). Middle: sagittal view of the T1-weighted MRI with isosurfaces of shortest ...