TODO

[emiwar]

• [x] Deconvolution (OASIS?)
• [x] Spatial restriction
• [x] Auto-noise estimation / dual problem
• [x] Auto time constant estimation

[emiwar]

For OASIS: can one use kurtosis for oasis? Two possible ideas: 1) Tune $\lambda$ so that the residual has a kurtosis of 3 instead of a norm of $\sigma^2 T$. 2) Change OASIS to directly optimize kurtosis=3 (quite complicated but would be very elegant)

Another question is whether to use the residual of each pixels in the ROI, not the trace of the neuron, to calculate the kurtosis. This should give a more accurate estimate since the noise in each pixel should be roughly independent.