v0.1.7 Improved length measurements, speed, and output

[hansenjn]

News

• Improved skeletonization for better length estimates, intensity profiles, orientation vectors, ...: Previous versions of CiliaQ did not very accurately measure the length of cilia in 2D images with low pixel lengths (e.g., <0.05 µm / px) (see also issue Curved cilia give wrong cilia length #13). The reason for this problem was in the implementation of the length measurements. For more explanations see below. In this new version, length measurements were largely improved, for all images and in particular for 2D images with low pixel lengths. However, as a consequence, length measurements obtained by this version slightly differ compared to measurements conducted with previous versions.

---

Do not mix any skeleton-related data (e.g., length, bending index, intensity profiles) obtained with this version with data obtained with previous versions, or vice versa. Otherwise, your merged data set may be biased by the difference in methodology across versions!

---

• For non-timelapse images, the generated 3D files are now stored as a .png file instead of .tif to reduce the storage space required by the output files.
• Garbage collection was improved (less garbage collection calls) to further speed up the program.

More details about the length measurements

What happened in previous versions:

To well estimate the length of very small, sub-diffraction-limited objects such as cilia, CiliaQ upscales (3-fold) and blurs each cilium object before skeletonizing it in order to determine the cilium length. To conduct these image processing steps in an efficient manner, CiliaQ generates a new small image for each cilium that is just a little bit wider than the original cilium, reducing the size of the image that needs to be processed and thereby, speeding up the process. Since CiliaQ was originally developed for 3D images only, it was implemented that a 3D image with additional empty planes on top and bottom was generated in this step. Thus, skeletonization for 2D images was also performed in 3D. Since the images was upscaled in this process, also 2D images were scaled in 3D and extra planes were generated between the empty planes on top and bottom and the original 2D plane to upscale the image in 3D. In the following the Gaussian blur was applied. Since extra space in x and y directions was a fixed number of a few pixels in these images, bluring, depending on the size of the blur, could result in widening the objects until the image boundaries, when the image size was to small for a large blur.

All these steps could unpredictably introducing small biases into the length measurements of in particular 2D cilia. Since 3D images were generated from the 2D cilia, an elongation of the skeleton, extra branches, or shortcuts could occur. In addition effects at the edges due to a "too small" image for a "too large" blur could slightly elongate or shorten detected cilia lengths. These inaccuracies due to edge effects particular affect the measurements of very short cilia and not so much long cilia. In long cilia a slightly longer length measurement would not largely change the cilium size in relative terms (changes may be just a few percent compared to the length of the cilium). In small cilia, a slightly longer length measurement would drastically change the cilium length in relation to to the small cilium size (could change the length measurement by 100-200%).

What has been changed in the length measurements for this version:

The new version of CiliaQ does no longer determine the skeleton in 3D for 2D images. Instead all steps are performed in 2D only. Besides that, the new version of CiliaQ adds sufficient extra space in all directions when generating the individual cilia images for length measurements, avoiding edge effects as described above.
Both of these improvements result in a much more accurate length detection compared to previous versions.

---

This discussion was created from the release v0.1.7 Improved length measurements, speed, and output.