@James-cole issue 3 is a duplicate of issue 2 (albeit hinting at the solution). Therefore, this PR should close 3 issues.
@james-cole this PR should close issue 3. @samanemati89 and I tested the patch suggested by @JohnAshburner on the scans from 61 adults. With this patch, r7487 and the recent version generated age estimates that were extremely highly correlated with each other, and both highly predictive of actual age. Numerically, the more recent version did a tiny bit better job. This difference was no where near statistically significant. However, the direction is nice because it does not suggest a larger sample size would be likely to reveal a preference for obsolete releases. Based on these findings, we feel this issue could be closed if the following changes are incorporated.
Great work @neurolabusc ! Really appreciate your time in sorting these issues. Also, it's reassuring to know that the model's prediction are generally accurate in your data.