Papers:
Cell-X (Zurich): http://www.csb.ethz.ch/tools/software/cellx.html
- Approach written in MATLAB, for use in cell segmentation
- Uses CLAHE to threshold, then uses unique membrane dynamics to examine cell boundaries.
- https://academic.oup.com/bioinformatics/article-lookup/doi/10.1093/bioinformatics/btu302
Large-scale automated identification of mouse brain cells in confocal light sheet microscopy images: https://academic.oup.com/bioinformatics/article/30/17/i587/201138/Large-scale-automated-identification-of-mouse
- Approach:
- Substack entire thing into WxHxD pieces
- Make sure substacks overlap (some dimension M) such as to prevent borders from improperly being trained (make sure all cells fall within WxHxD? need more reading)
- Two inputs: L, S (each point intensity represented as (x, y, z))
- L - set of voxels whose intensities exceed background intensity
- S - specially selected subset to run mean-shift algorithm
- Mean shift algorithm
- The classic mean shift algorithm would start from all available data points, place a kernel on each of them and shift each point toward the mean value computed as the kernel-weighted average of the data.
- In this variant, they improve both its running time and its statistical precision by starting from a carefully chosen set of seeds S
- Thresholding
- Found background/saturated areas and removed as basic filters.
- Supervised semantic deconvolution
- Above step works well on bright images. Darker somas might be missed.
- To improve over this intrinsic difficulty, they carried out a preprocessing stage by applying a non-linear filter trained to boost weak somata and decrease the voxel intensities in non-soma regions.
Gradient Flow: https://www.ncbi.nlm.nih.gov/pmc/articles/PMC4578315/
- Approach is to find "gradient" of background (ie - derivatives w.r.t X, Y, Z) and set them to zero
In [ ]:
In [ ]:
In [ ]:
In [ ]: