Skip to main content
. Author manuscript; available in PMC: 2022 Dec 30.
Published in final edited form as: Neuroimage. 2022 Feb 3;251:118933. doi: 10.1016/j.neuroimage.2022.118933

Table 15.

Compiled results of Fig. 12 - Part 2: Comparison of fixed-resolution networks and multi-resolution FastSurferVINN. State-of-the-art FastSurferCNN and optimized FastSurferCNN* trained on 120 mm scans are compared to FastSurferVINN trained on a mixed set (60 high-resolution and 60 1.0 mm scans). Note, that the scans are equivalent between the training sets and only differ in their resolution. Dice Similarity Coefficient (DSC) and average surface distance (ASD) are reported for six 1.0 mm datasets.

Networks FastSurferCNN (n = 120) FastSurferCNN* (n = 120) FastSurferVINN (n = 120)
DSC ABIDE-I 83.33 (± 4.29) 84.55 (± 3.46) 85.302.88)
ADNI 86.99 (± 2.15) 87.59 (± 2.09) 87.941.91)
IXI 84.46 (± 4.61) 85.22 (± 5.05) 86.143.70)
LA5C 88.3 (± 1.40) 88.84 (± 1.33) 89.251.35)
OASIS1 88 (± 1.38) 88.53 (± 1.37) 88.891.39)
OASIS2 87.36 (± 1.53) 87.81 (± 1.51) 88.181.26)
ASD ABIDE-I 0.441 (± 0.140) 0.416 ( ± 0.129) 0.4020.125)
ADNI 0.315 (± 0.094) 0.294 ( ± 0.086) 0.2810.079)
IXI 0.381 (± 0.128) 0.367 (± 0.148) 0.3320.106)
LA5C 0.273 (± 0.045) 0.256 (± 0.044) 0.2410.043)
OASIS1 0.278 (± 0.062) 0.264 ( ± 0.063) 0.2550.066)
OASIS2 0.29 (± 0.048) 0.276 ( ± 0.048) 0.2620.039)