In this study, we present a comprehensive evaluation of two brain segmentation methods using three atlas databases. We present detailed descriptive data comparing three commonly used atlas databases and show that the databases differ in quality.
Both segmentation methods reliably identify known abnormalities in each patient group; FreeSurfer separated better between patients and healthy controls in the AD and left HS datasets, whereas MAPER performed better for the right HS dataset.
CVs for region volumes for the HM and DKT40 atlas databases were similar, while the MGC2012 atlas showed a higher mean CV across all regions. The HM atlas database had the largest range of SVR (max/min = 9.0), followed by MGC2012 (5.8), and DKT40 (2.1). This has implications in interpreting the overlap between automatic and manual segmentations, because overlap tends to decrease in regions with higher…
