2p imaging

[kjohnsen]

• [ ] limit image to 2D plane (I think. verify)
• [ ] potential cross-talk via low-level background light level at specified wavelength (more realistic would be tiny blips of higher intensity light corresponding to scanning, but we don't want to simulate at extreme temporal resolution)
• [ ] separate out the indicator model to be injected separately. Hopefully straightforward kinetics models exist for voltage and Calcium indicators. Or in the case of Ca++ maybe don't simulate ODE, but compute closed-form convolutions of spikes. And noise.
• [ ] output ΔF/F traces for each ROI
• [ ] (maybe) generate images from output data (show ROIs from traces, add noise to background)

[kjohnsen]

Talked with Adam Charles at Cosyne about approximating optical recording for point neurons. He says NAOMI has a function to get calcium traces from spikes. It involves adding Gaussian and Poisson noise (or maybe that's for generating images?). Maybe the distance of the neuron from the plane of imaging could be implemented as scaling with 1/d outside of a certain range. Similar to what I do with spike detection probability. NAOMI runs an ODE to generate Ca++ traces--perfect for Brian. I still have questions about voltage imaging and holographic stimulation, if he might be able to help.

[kjohnsen]

Notes from meeting with Adam Charles about Ca/v imaging, 5/31/2023

• Different levels of realism for volage:
  • Just generate traces without membrane-bound indicators, tune laser power down to approximate low fluorscence levels
  • More realistic would be membrane-bound GEVI

• generateTimeTraces gives you no-noise as well as noisy outputs. Noisy outputs are from video, true ROIs.

• Noise can depend on location (in out of plane), or expression, or occlusion (blood vessels),

• If I want uniform noise approximation, electronics Gaussian noise, P(N_t) noise, SNR from illumination

  • Simpler model: Poisson-Gaussian. Poisson dist on fluorescence (N_t, photon count at time t), fed into Gaussian where mu and std are affine functions of N_t. Decent, but need to fit 4 params. Not clear how to apply to voltage data.
  • Could fit to experimental data via MLE (don't need ground truth).
  • Could be interesting to see how params vary for different datasets, and for calcium versus voltage
  • Unpublished, but could dig up if needed

• Run online? Could simulate 1 frame at a time. Would need spikes from the past few milliseconds

  • He thinks it's possible. NAOMI is all causal.

[kjohnsen]

Forget about plane visualization: just highlight neurons that are being imaged, maybe with a different marker than used for all neurons

[kjohnsen]

closed with #39