bencardoen / smlmtools.jl Goto Github PK

Only x/y/z is saved, not the remaining metadata fields, change flow so extra data is saved as well

Adding Plots as a dependency is cumbersome, e.g. right now it's causing CCI to fail builds because of an arcane error in FFMPEG.jl, when CCI runs as root.
For now, remove all plotting functionality and deps on Plots.

Alternatively, save the plotted data arrays to csv, so users can plot them later.

This also dramatically reduces dependencies

Smlmvis has package conflicts that make deployment painful, due to pillow/vtk conflicting.
The current package works, but it's not very stable.
Instead, use https://juliapackages.com/p/writevtk to write to VTU, and use the new smlmvis version without vtk support

Add output to VTU (point cloud format that can be visualized in ParaView)

Detect up to N fiducials

• User specifies how many fiducials to expect
• user specifies tolerance in nm of maximum distance edge to edge between fiducials
• detect_bead --> do not rely on overlap
• for each channel
  • detect N bead candidates
  • find closest pair
  • if distance closest < threshold, proceed, else quit

dStorm data can have long Z tails, add functionality to filter on the focal plane in Z, and allow only a minimum precision

For channel 1 the current CSV point cloud output is not stored correctly, it's saving the fiducial correction, not the actual point cloud.

The current bead detection uses the observation that beads emit almost always, so create hyper-dense regions.
However, it is not unusual for beads to become detached, or fail, in which case they do not have emissions for a large span of time.
The current detection method could, in the worst case, using an unstable bead, and have incorrect tracking information.
The current criteria ranks candidates by their emission density (top k).

Proposed correction

• For each detected bead, compute the temporal presence of that bead (>90% of frames)
• If a candidate does not have emissions for a configurable amount of time, discard it

Export in CSV SRN Format

If multiple beads are present, and they do not overlap (~ 50nm or so), the tracking code fails to detect.
The code already picks up multiple beads.

• Detect distances between beads
• Pair nearest
• For this pair, if distance < threshold, proceed, else quit

Add support for reading 2D Thunderstorm CSV data