Some of the cells from NeuroMorpho.org manifest z-axis jitter. This artifact can add quite a bit extra cable length to each cell. As far as I remember from Neuron course, this artifact is caused by the fact that gears on experimental set up are not connecting perfectly, and therefore allow to turn slightly without changing the focus. However, I do not remember what is the strategy for dealing with these kind of reconstructions? Should I discard cells with z-axis jitter? It would be best to still use these kind of cells.
Thanks
Dealing with z-axis jitter in reconstructed cells
-
ted
- Site Admin
- Posts: 6300
- Joined: Wed May 18, 2005 4:50 pm
- Location: Yale University School of Medicine
- Contact:
Re: Dealing with z-axis jitter in reconstructed cells
The Import3D tool
NEURON Main Menu / Tools / Miscellaneous / Import3D
offers several methods for trying to automatically repair z axis artifacts. After importing the morphology file, click on the "Show Points" button to eliminate all the blue boxes, then click on "Rotate 45deg about y axis" twice to get a side view. Next click on Import3D's "Edit" button for a panel of buttons that allow you to select from a menu of strategies for dealing with z axis errors.
That said, I wouldn't waste my time on such data. First problem is: how can you tell that any particular point is affected? Set a threshold for z axis jumps? Risks false negatives (missing artifactual jumps that are less than the threshold) and false positives (neurite trajectories that genuinely deviate from the xy plane may be mistaken for a series of artifactual jumps). Even if you can accurately identify artifactual jumps, what do you do to fix them? Reduce delta z at each affected point to 0? Infer delta z from the delta z values of neighboring points, e.g. by linear interpolation or spline, or probability distribution, or 3 point median filter? Why deliberately use data that one knows is faulty and cannot be fixed in a way that is "guaranteed" to work?
NEURON Main Menu / Tools / Miscellaneous / Import3D
offers several methods for trying to automatically repair z axis artifacts. After importing the morphology file, click on the "Show Points" button to eliminate all the blue boxes, then click on "Rotate 45deg about y axis" twice to get a side view. Next click on Import3D's "Edit" button for a panel of buttons that allow you to select from a menu of strategies for dealing with z axis errors.
That said, I wouldn't waste my time on such data. First problem is: how can you tell that any particular point is affected? Set a threshold for z axis jumps? Risks false negatives (missing artifactual jumps that are less than the threshold) and false positives (neurite trajectories that genuinely deviate from the xy plane may be mistaken for a series of artifactual jumps). Even if you can accurately identify artifactual jumps, what do you do to fix them? Reduce delta z at each affected point to 0? Infer delta z from the delta z values of neighboring points, e.g. by linear interpolation or spline, or probability distribution, or 3 point median filter? Why deliberately use data that one knows is faulty and cannot be fixed in a way that is "guaranteed" to work?