Manual ROI drawing tool - Shift in ellipsoid centroids after trace extraction: False ROI positions and activity traces?

[TomRiddleVorlost]

Dear suite2p creators and suite2p team,

First and foremost, thank you for all the work invested into the suite2p software. It really brings awesome features and a conveniant GUI to the table, complementing, and in some domains outperforming existing calcium imaging analysis tools.

The issue I am struggling with at the moment concerns the manual ROI drawing tool and consists of two parts:

1. In my hands, manual ROI drawing (to be more precise the extraction of the traces step after ROI drawing) only works directly after suite2p has run (using the GUI in Anaconda). While I can draw elliptic ROIs directly after suite2p run-through (before closing the GUI), when I reload the 'stats' file from the 'combined' folder of an experiment that has been processed by suite2p days/weeks before and in which no manual ROI drawing has been performed yet, I am able to draw manual ROIs as usual, but the extraction of the traces fails (despite having had the ops settings: delete_bin = False; combined = True; keep_movie_raw = True; reg_tif = True). I am using a single channel recording 512/1024 pixels at 3 Hz processed in 200 frame chunks). When trying to extract the traces of the drawn ROIs "ValueError: need to provide number of frames n_frames when writing file" occurs (see panel A in the Issue Illustration Figure attached).

Is n_frames somehow cached directly after running suite2p with the gui and not anymore when reloading stats files later? How can I provide the software with this information when reloading stats files to (re)perform manual ROI drawing?

2. This is the more confusing part to me: I deselected all functionally detected ROIs/cells of a measurement by suite2p by adjusting the cell classifier to 0.999. This "emptied" the (selected left panel) "cells" to the deselected panel. I then drew big, elliptic ROIs of entired brain regions (see panel B in illustration) to only get the average signal of individual regions and not that of single neurons (see panel B in illustration). When hitting the button to extract the traces, I noticed a curious shift of the individual "centroids"(cX in panel B,C) of the elliptic ROIs while the outlines were still the ones I drew (panel B). This shift of the little crosses (centroids) did not seem to be systematic, at least to my understanding, since centroids of individual ellipsoids deviated more or less along both x and y axes (panel B). Afterwards I saved and exited the Manual drawing menu and closed the software.

When reloading the updated stats file, surprisingly it seemed as if the manually drawn ellipses/ROIs were now shifted as well (original position: dashed outline; new position: color filled ellipses, panel C). To me it seemed as if the "new" ellipses were constructed around the falsely shifted centroids, leading to a wrong ROI in the GUI. In panel C you can observe an image-stack of 9 planes in which multiple elliptic ROIs have shifted dramatically (B original positions and wrong centroids; C shifted positions based on wrong centroids?). One ROI (roi0) even disappeared completely. Maybe I am missing
something, but I could also not find a systematic error related to the ellipses size or position (panel C). The launching point of the ROI (when hitting the "add roi" button) at least did not change things, indepdendent where the ROIs were launched, the centroid would not occur at the "launch site" but somewhere nearby the ellipsoid ROI (inside and outside of the outline both occurred).

Has this ever happened to anyone of you? DO you have any advice what could be going wrong here? I only noticed it, drawing big ROIs, however it also held true for small neuron-sized ROIs (panel E).

My main concern is, are the traces that I extracted even the ones from the original elliptic ROI I drew, or are they the traces of the newly generated "wrong" ellipsoid ROI (which I would assume)?

I hope I made myself understandable and that you guys can help me out, I am a bit puzzled... As a heads up, I have the most recent version of suite2p (updated 2 weeks ago) and am using the GUI launched in Anaconda on a Windows device.

Thank you very much in advance for any help,

Tom Offner

[Anantu-Sunil]

Hi,

I'm trying to extract traces from olfactory bulb glomeruli, and manual labelling is more or less the only way to go about it. But I'm facing the same problem of manually labelled ROIs showing shifted centroids after extraction.

I tried with and without non-rigid registration, but the issue persists. It would be great if someone could look into this.

Thank you,
Anantu Sunil