Registration Workflow and Best Practices
This page explains the practical workflow experienced users follow to get good atlas registrations quickly.
Start with the automatic outline points
Begin with the four points created by the automatic registration.
These points are placed on the brain outline.
First correct them so they match the visible outline of the 4x brain slice as well as possible.
This gives you the initial whole-slice fit before you refine anything else.
Add points where they improve the fit
After correcting the outline points, add new points only where they clearly improve the registration.
Focus on visible anatomical landmarks that help the atlas match the slice better.
Spread points across the relevant area instead of stacking many points in one small place.
Choose landmarks that are clear and reliable
Use landmarks that are easy to recognize in both the atlas and the slice.
Good choices are strong outer contours, characteristic bends, and clearly visible internal boundaries.
Avoid weak or ambiguous features that are hard to interpret consistently.
Cycle through stainings when useful
Different stainings reveal different anatomical structure.
Some stainings make relevant regions easier to see, while others may not help much.
Switch channels when needed to judge a boundary or landmark more confidently.
Remove or fix unhelpful points
If a point is clearly wrong or starts pulling the fit in the wrong direction, move it or delete it.
A smaller number of good points is better than a larger number of poor ones.
Point quality matters more than point count by itself.
Know when a slice is done
Stop when the atlas matches the visible anatomy well in the area that matters for your analysis.
Only adjust features that can actually be seen in the slice.
It is normal that some atlas boundaries appear jagged, because the atlas is defined on pixels.
Those jagged lines are an expected atlas limitation and should not be corrected.
In practice, around 10 to 20 well-placed points is often enough for a very good result, but the exact number depends on the slice.
Final note
The goal is not to force every line to look perfect everywhere.
The goal is a trustworthy registration in the visible anatomy that matters for downstream analysis.
