Cleaning up optical crosstalk between simultaneously imaged planes in a two-photon mesoscope
Year: 2026
Type: article
Abstract: Two-photon mesoscopes let us record from a lot of neurons at once, in the Allen Brain Observatory and many other projects across the Institute. One trick for recording even more neurons is to split the laser and image two depths at the same time. It comes with a subtle problem: the two planes bleed into each other. A faint copy of one plane shows up as "ghost" cells in the other. Left uncorrected, those ghosts look like real neurons. Just as important, the bleed-through contaminates the activity traces (ΔF/F) extracted from genuine ROIs, degrading real cells — not only adding spurious ones. This post walks through the problem and the correction we run in production: what the crosstalk is and where it comes from, why the obvious fixes don't work, how our estimator works (minimizing mutual information between the paired planes from the bright ROIs), and how well it does — measured against real ground-truth recordings, against simulations, and against two off-the-shelf alternatives (naive ICA and suite3d). I'll also cover the quality-control tooling we built to catch the rare failures.
Author: Kim, Jinho
Institutions: Allen Institute, Allen Institute for Neural Dynamics
Language: English
Subfield: Biophysics
Domain: Life Sciences
Open Access status: green
OpenAlex ID: W7172205524