Hi, couple of things regarding this:
First, the algorithm of BDP for automatic masking did change in 15a as compared to 14a, which was also based on feedback from BDP users. We tried to make it more robust to variety to datasets in our test-sets. However, it is possible that the performance for your dataset is not optimal with this change. If you can send us a sample dataset, on our support email address, then we can include it in our test-set.
Having said that, if you have masks from 14a run then you can directly use the generated masks with any version of BDP. It should work without any problem.
Also, in case you are editing masks manually, you do not need to make very precise masks. You can just use thresholding in mask-tool – just make sure you are not missing big chunks of brain.