Calculating Volumes for Segmented Regions

[AndreiRoibu]

Is your feature request related to a problem? Please describe. We need to build confidence that brains segmented with P4B produce similar results to those segmented with the original QuickNAT. We also need to check if volume predictions over periodically scanned patients are consistent with clinical observations (such as the hypothalamus staying constant in size vs other regions)

Describe the solution you'd like

1. A clear and statistically viable comparison between QuickNAT and P4B, showing no statistically significant difference between the two.
2. A clear indication of P4B being able to capture clinically significant difference in periodical scans of the same patients.

Additional context Initial decisions to be made:

• Number of samples for statistical significance
• Statistical tests
• Manner of presentation

[AndreiRoibu]

In the original paper (here), the author uses the ADNI-29 database and the following tests:

• for segmentation consistency, it uses the volume distance between the estimated volumes of different structures, a metric which indicates the error in volume estimation after segmentation, where Va is the actual volume and Ve is the estimated volume of a given structure. dv(Va, Ve) = 2 * abs(Va-Ve) / (Va+Ve)
• Hedges's g and Glass Delta variants customized for small samples were used for reporting the effect sizes for the group analysis when comparing AD patients with controls.
• volume estimates were also normalized for various segmentation methodologies using a linear model, with the regression co-efficients and p-values being reported.

The above should try and be reproduced to compare P4B with the original QuickNat.