Q&A: Most of my analyzed cilia show more than one branch what does it mean?
Most of my analyzed cilia show more than one branch - does this mean tracking of the cilia is incorrect?
Question: "Most of my analyzed cilia show more than one branch - what does it mean? Do I need to be concerned about the quality of the reconstruction and output data? I observed also that when applying Canny3D as a segmentation method I get much less side branches than with an intensity-threshold-based method."
Answer provided on 21st of March 2021.
Generally, when for many cilia much more than one branch is indicated this does not necessarily mean that the detection of the cilium is incorrect. It may just serve as an alert signal that should remind you to check visually whether segmentation & detection is fine.
It can also happen that multiple branches are detected for a cilium when you have noisy images and thus noisy ciliary reconstructions. For example, I recently analyzed some images where the signal to noise ratio of cilia was not good - in turn, I got cilia reconstructions that did not show a very smooth surface. Such detected cilia from noisy images may be indicated with more than one branch in the results file, because, in addition to the ciliary centerline, such side arms from the small inhomogeneities (looks like "pocks" in the image below) on the ciliary surfaces are appearing in the ciliary skeleton.
Noisy images (left) may result in a noisy cilium detection (right, 3D rendering obtained with FIJI's VolumeViewer implemented in CiliaQ) - original image provided by (c) J.N.Hansen
You will of course have much less of these inhomogeneities if you use the Canny3D technique because it includes a blur. So one option could be that you add a small Gaussian Blur to the image (you can select this in CiliaQ Preparator, also when not using Canny3D) to reduce side branches. See also Fig. 1G in the CiliaQ Publication, showing for an exemplary image how the occurence of side branches is depending on the applied Gaussian Blur. Still as long as you can see that the skeleton correctly picks up the cilium (stretches across the center of the cilium) the side branches do not matter. E.g. for the cilium shown above, CiliaQ revealed this skeleton, which nicely maps the center line only:
Detected centerline of the skeleton (3D rendering obtained with FIJI's VolumeViewer implemented in CiliaQ) - original image provided by (c) J.N.Hansen
As it is implemented in CiliaQ that the only the "largest shortest path" in the skeletal tree is considered, all other branches are and, in this case, were neglected. So if the detected centerline for the cilium matches its shape the detection is fine, even if a high number of branches is reported.
However in cases like the following, you must definitely improve the analysis, because the skeleton does not pick up the correct cilia shape, but follows side branches originating from the noise.
Noisy images (left) may result in a noisy cilium detection (center), which in turn may also result in incorrect detection of the cilia centerline (right) (3D renderings obtained with FIJI's VolumeViewer implemented in CiliaQ) - original image provided by (c) J.N.Hansen
To sum up, a high amount of side branches is an indicator for either a suboptimal segmentation method or high image noise. Thus, you should carefully scrutinize the detection of the centerline and the reconstruction of the cilium in the output images. However, as long as the ciliary centerline is correctly picked up, you can also accept detection of side branches when looking, e.g., into the ciliary length. In any case, it may be good to test whether you can optimize the segmentation method to reduce side branches but that may not always succeed.
Should you not look into skeleton based parameters ("centerline" based parameters, cilia length, cilia orientation), you do not need to care about branches or any skeleton results. But - caveat - of course you should pay attention to the fact that such extra appendages caused by noise may deliver also values to other output parameters like the volume or the average intensity. As long as the noise is equal in all your images that may not be so problematic because the bias would be equal. If, however, in only few images the high image noise causes these extra-appendages in the ciliary reconstruction, you may consider that these may influence the results for the respective images.