Skip to main content
. 2019 Oct 21;41(2):309–327. doi: 10.1002/hbm.24803

Table 5.

Comparison of four of our CNN‐based segmentation methods with previously reported results (Zandifar et al., 2017) for the segmentation of the left and right hippocampi in the ADNI dataset. Each table cell reports the mean Dice coefficient (SD) as a percentage on top and the mean MHD (SD), in millimeters, on bottom

Left Right Both
CNN‐B 90.7 (2.3) 89.8 (3.2) 90.2 (2.8)
0.25 (0.07) 0.29 (0.12) 0.27 (0.10)
CNN‐SP 91.5 (1.9) 91.6 (2.1) 91.5 (2.0)
0.23 (0.07) 0.23 (0.09) 0.23 (0.08)
CNN‐SP‐D 92.0 (1.6) 91.8 (2.2) 91.9 (1.9)
0.21 (0.05) 0.23 (0.08) 0.22 (0.07)
CNN‐SP‐D + DA 92.0 (2.0) 92.2 (2.1) 92.1 (2.0)
0.22 (0.07) 0.21 (0.07) 0.22 (0.08)
ANIMAL 86.3 (2.6) 85.9 (3.0) 86.1 (2.8)
0.40 (0.07) 0.42 (0.09) 0.41 (0.08)
ANIMAL + EC 86.5 (2.4) 86.2 (3.0) 86.4 (2.7)
0.43 (0.08) 0.44 (0.09) 0.44 (0.08)
FreeSurfer 75.8 (4.7) 75.6 (4.8) 75.7 (4.7)
0.94 (0.27) 0.98 (0.24) 0.96 (0.26)
FreeSurfer + EC 85.9 (3.3) 86.3 (3.1) 86.1 (3.2)
0.44 (0.13) 0.42 (0.09) 0.43 (0.11)
PBS 87.5 (2.5) 87.3 (3.6) 87.4 (3.1)
0.40 (0.09) 0.40 (0.13) 0.40 (0.11)
PBS + EC 88.2 (2.5) 88.2 (3.6) 88.2 (3.1)
0.39 (0.08) 0.40 (0.14) 0.39 (0.11)
PBS + NLR 88.3 (2.2) 88.0 (3.2) 88.1 (2.7)
0.39 (0.07) 0.39 (0.12) 0.39 (0.10)
PBS + NLR + EC 89.1 (2.6) 88.9 (3.1) 89.0 (2.6)
0.37 (0.06) 0.37 (0.12) 0.37 (0.10)

Note: The top performing method is emboldened in each column.

Abbreviations: CNN, convolutional neural network; CNN‐B, baseline CNN; CNN‐SP, CNN with spatial priors; EC, error correction, NLR, nonlinear registration, PBS, patch‐based segmentation.