Hello

I’m not familiar with this segmentation tool and I’m trying to perform an extra cranial facial nerve segmentation on MRI 3D T2W sequences. Can some one help me with the steps to follow to obtain a linear segmentation?

Probably the easiest way of segmenting facial nerves is to use “Draw tube” effect (provided by SegmentEditorExtraEffects extension) in Segment Editor module. If you don’t need segmentation, but you just want to mark a curve then you can simply place a curve.

Thanks a lot, I succeeded with these extensions in segmenting the facial nerve.

My study actually consists of comparing the segmentation of extra cranial facial nerve made by a junior radiologist and a senior radiologist specialized in head and neck.

I have issues when using the RT slicer module to compute haudsorff distance and Dice coefficient. The program always shuts down when trying to compute it or returns zero for all metrics.

I also want to ask if it s possible to compare stored segmentations later when needed.

Dice coefficient is not applicable to thin or long structures. Hausdorff distance only gives information on the position of the largest deviation, so it does not give you very detailed information either.

If you only need to compare the centerline curve of the nerve (no need to compare radius) then you can write a short Python code snippet that computes detailed statistics (such as mean max error, and 50th, 75th, 95th percentiles; errors in different segments of the nerve, …).

The script could first resample the assessed curve at equal distances (using `ResampleCurveWorld`

method) and then for each control point position get the distance from the ground truth curve (using `GetClosestPointPositionAlongCurveWorld`

method), and add all the distances into a numpy array. You can then use `mean`

and `max`

numpy functions to get the mean error, and `percentile`

numpy function to get percentiles. Probably about 15 lines of Python code in total.

Thanks for your time and patience. I’ll try to find someone to write the script since it 's too complicated for me. Meanwhile can you check this article ? They used exactely the same technique and coefficient I mentionned using 3d slicer." MR Imaging of the Extracranial Facial Nerve with the CISS Sequence" J.P. Guenette, 2019

Regardless of this paper passing peer review, Dice Coefficient metric is not suitable for comparing thin tubular objects. Average Hausdorff distance is meaningful, it is just not very sensitive metric. In addition to average, higher percentiles and/or computing the average for parts of the nerve would have been more informative.

I’ve just tested Segment Comparison module in Slicer-4.11.20210226 on Windows and it worked well. What Slicer version did you use? Can you attach a screenshot of your Segment Comparison module? Do you see any error message in the application log? (you can open the log of the previous session in menu: Help / Report a bug)