Picasso: Tools for Painting Super-Resolution Images

Picasso performs 3D calibration by fitting a six-degree polynomial to the mean $x$ and $y$ spot widths ($\sigma_x$, $\sigma_y$) across different $z$ positions.

When calibration finishes, Picasso generates a six-panel diagnostic figure (saved as a .png next to the .yaml calibration file) to verify quality:

• Mean spot width/height vs stage position: Shows the measured widths and the fitted polynomials. Picasso shifts the axis so fits meet at $z=0$.
• Spot width vs spot height: A scatter plot of all kept localizations. A narrow band following the curve indicates good calibration; a wide cloud suggests field-dependent PSF or a tilted stage.
• Spot width/height vs estimated z: Similar to the first plot, but showing individual beads.
• Estimated z vs stage position: Shows recovered $z$ against known stage position. Points should lie on the diagonal. The range where they follow the diagonal is the usable depth.
• Deviation to true position: A histogram of estimated z - stage position. It should be single-peaked and centered on 0.
• Mean z precision vs stage position: The RMS deviation per $z$ step. Note that field-dependent PSF or a tilted stage can impact these values.

Note: These diagnostics report the self-consistency of the calibration beads, not necessarily the performance on dim single molecules.

When using multiple cameras, Picasso uses the config.yaml to organize the selection UI and automate camera matching.

Camera Selection Logic

• Automatic Matching: Picasso attempts to match the camera name in the config with the name found in the Micro-Manager metadata. If no match is found, the first camera listed in the config is selected.
• Alphabetical Order: By default, configured cameras are displayed in alphabetical order in the dropdown menu.

Setting Camera Priorities

To prevent the dropdown menu from becoming cluttered, you can use the CameraPriority key. This allows you to pin specific cameras to the top of the list.

• The CameraPriority list must contain names that match entries in the Cameras section.
• Listed cameras appear at the top; all others follow in alphabetical order.

Cameras:
  Camera1:
  Camera2:
  Camera3:

CameraPriority:
  - Camera3
  - Camera1

When performing multichannel 2D spherical Gaussian fitting, you can choose how photon counts and backgrounds are handled across channels.

Unlinked (Default)

Each channel fits its own photon count and background. Only x, y, and width are shared. This is appropriate for channels with different spectral throughput or uneven beam splitters.

This mode adds the following columns per channel c (supports 2 to 6 channels):

• photons_ch<c>: The channel's specific photon count.
• bg_ch<c>: The channel's specific background.
• rel_photons_ch<c>: The channel's share of the total photons (values sum to 1 per localization).

Linked

One photon count and one background are shared across all channels. This assumes the channels are balanced (e.g., equally split, redundant channels).

Note: In both modes, the photons and bg values represent the totals across all channels, making them directly comparable to spline fits.

The Temporal median filter (in the Identification group) helps remove uneven backgrounds like out-of-focus haze or non-uniform illumination by subtracting a rolling per-pixel median over a window of frames.

Key Details:

• Identification Only: The filter is applied only during the spot identification phase. Spots are always cut out and fitted using the raw movie data. Therefore, photon counts and localization precision are not affected by the filter.
• Window (frames): Sets the window size for the median. A default of 51 is recommended. It must be long enough for emitters to be dark most of the time, but short enough to follow background drifts.
• Tuning: Because subtracting background changes the local gradient scale, you must re-tune the Min. net gradient parameter after enabling the filter. Use the Preview mode to verify.
• Restriction: The filter is not applied during 3D or experimental PSF calibration, as calibration beads are static and would be subtracted by the filter.

When one channel has very low signal, identifying channels separately may fail to detect molecules. The Sum of channels mode allows identification in a combined signal.

How it works

1. All channels are mapped onto the reference channel using the transforms from the loaded calibration.
2. The channels are summed in photons.
3. Spots are identified in this summed image.
4. The resulting detections are then used in a joint fit (the fit itself does not link across channels, as the detections are already cross-channel consensus).

Requirements and Tips

• Registration is mandatory: Channels must be registered (via a loaded calibration or prior identification) before they can be summed. If registration is off, the sum will be smeared.
• Re-tuning: You must re-tune the minimum net gradient because the summed signal scale differs from single-channel scales.
• Re-alignment: If the calibration registration is poor, run Calibration > Re-align channels (current signal) before attempting to sum.

API Access

This mode can be accessed via script using:

• picasso.localize.identify_multichannel_sum for identification.
• picasso.localize.SummedChannelsMovie for the summed view.