Apply Sequential Monte Carlo estimation alg to mouselight data

[tathey1]

Here is a paper with an interesting approach to neuron reconstruction.

To what extent can it be used/is it effective on our data?

[shreyasingh1]

I'm interested in exploring this method more, and if feasible, running it on the mouselight data. Their concept of using sequential Monte Carlo estimation down the tree structure of a neuron to develop probabilistically independent branch estimates, and then combining them into one full neuron tracing model, seems sound and plausible.

However, their method assumes each image stack has either one soma or no soma, which may not be true for our data. Additionally, the method requires accurate knowledge of where the soma is so it can set a starting point for the sequential Monte Carlo estimation. This may be something we need to apply soma detection code to or estimate manually for some time. This method has been applied to BigNeuron datasets -- we should note that BigNeuron does not currently contain photoelectron microscopy data, so this algorithm's compatibility with the Mouselight data type is not guaranteed.

[bvarjavand]

Our ground truth data is a list of coordinates tracing a neuron, with the first vertex always at the soma. You can visualize the data with neuron ids 2 and 7 loaded.

[alishakodibagkar]

I attempted to address this issue by building the Vaa3D plugin for PNR (Probabilistic Neuron Reconstructor) from the source code provided in the paper.

Below are the challenges I faced:

• Unable to download version of Vaa3D beyond 3.100 on Mac - may result in incompatibility issues
• Switched to Windows - does not work on Mac
• Model requires old version of Qt that is incompatible with Mac
  • Qt 4.7.1
• Things to download: Qt 4.7.1, Boost
• Error:
  • “g++.exe: error: ........v3d_externalv3d_mainbasic_c_funv3d_message.cpp: No such file or directory, g++.exe: fatal error: no input files”
  • Unable to solve this error
• Some experience with C++ plugins/Vaa3D would be helpful