Find Spots

Spots are detected as local intensity maxima on the filtered images to create 3D point clouds. Spot positions are used later in register and call spots.

0: Auto Threshold

For each tile (\(t\))/round (\(r\))/channel (\(c\)) image, an automatic intensity threshold (shortened to auto_thresh) is computed. This is done by taking all pixels on a single, central z plane on the filtered image called \(I_{xy}\). The auto_thresh is

\[ \text{auto\_thresh}_{trc} = \text{n}^{\text{th}}\text{ percentile}(|I|_{trcxy})_{trc..} \times a \]

where \(|...|\) is the absolute value of each element separately. The median is computed over all x and y values to give a scalar that is a good lower bound estimate for the random noise amplitude. \(n\) is the auto_thresh_percentile (typically \(5\)). \(a\) is the auto_thresh_multiplier (typically \(160\)). Higher auto_thresh_multiplier makes spot detection stricter.

1: Spot Detection

For each tile/round/channel filtered image (\(I_{trcxyz}\)), a 3D point cloud is created out of all points where

\[ I_{trcxyz} > \text{auto\_thresh}_{trc} \]

2: Remove Duplicates

Some of these points will be nearby or adjacent, representing the same spot multiple times. To deal with this, points are removed. For each point, an ellipsoid region of x/y radius radius_xy (typically \(5\)) and z radius radius_z (typically \(2\)) is considered. If one or more points are found within the point's ellipsoid region, then only the point with the greatest intensity is kept. If a point is isolated, it is kept. The process is repeated for all points.

If too few spots are found for a tile/round/channel image, then a warning and/or error is raised to the user.

Diagnostics

Auto Threshold

The calculated auto thresholds can be seen from the notebook. For tile t, round r, channel c, the auto_thresh value is saved as float32 at

from coppafisher import Notebook

nb = Notebook("/path/to/notebook")
nb.find_spots.auto_thresh[t, r, c]

Detected Spots

A find spots viewer will help to understand the config parameters. Move sliders to adjust the find spots parameters and see the spot detection results. You can select any tile/round/channel combination. By default, the first tile's anchor round/channel is shown. To display the viewer, in the terminal

python -m coppafisher -fs /path/to/notebook

and open the link shown in the terminal on a modern browser to show and interact with the viewer. Press Ctrl + C in the terminal to close the viewer.

Spot detection is most crucial for the anchor round/channel images. So, it is recommended to configure find spots parameters based on one of these images.