Seeding ROIs for CNMF-E Pipeline

[Bilalgp]

For better support, please use the template below to submit your issue. When your issue gets resolved please remember to close it.

• Tell us a bit about your setup:
  1. Operating system (Linux/macOS/Windows):

Windows 10. RAM 64 GB. Intel i7.

2. Python version (3.x):

Python 3.6.12

3. Working environment (Python IDE/Jupyter Notebook/other): Jupyter Notebook (Jupyter Lab)

4. Which of the demo scripts you're using for your analysis (if applicable): CNMF-E Pipeline and Seeded ROIs demo pipeline

5. CaImAn version*: 1.8.7

• Describe the issue that you are experiencing Hello. I am interested in seeding the ROIs from one CNMF-E batch result into another. I tried following the seeded demo pipeline to the best of my ability. My code is saving cnm.estimates.A into the object Ain and then reloading it in a new CNMF-E batch with the following code:

    Ain_test = Ain.toarray() > 0
    cnm = cnmf.CNMF(n_processes=n_processes, dview = dview, params=opts, Ain = Ain_test)

However, for this to work, I need to set:

rf = None
only_init = False

But the CNMF-E pipeline requires only_init to be true. The same number of ROIs appear in the total components of the new batch but the contours seem a little different than the original shape in the plot_contours() function. Attached are images of the original batch and the new batch.

If I only set only_init to true, I get an error. If I set rf to any other value than None then the proper ROIs do not appear. The proper seeded ROIs do not appear either if I do the following (with rf set to None and only_init = False).

    Ain_test = Ain.toarray() > 0
    cnm = cnmf.CNMF(n_processes=n_processes, dview = dview, params=opts)
    cnm.estimates.A = Ain_test

Original Batch:

Seeded Batch:

Thanks.

[j-friedrich]

With the newest commit to the dev branch you can seed the CNMF-E method explicitly with a list/array of seed pixels, see the example below. centers could be the value of an earlier result cnm.estimates.center.

import caiman as cm
from caiman.source_extraction import cnmf

# data from https://www.dropbox.com/sh/x0xx6mxqg6llj3u/AADw9LobDktqLF4Eo9YSTewga?dl=0
fname = 'bma22_epm.mat'
Y = cm.load(fname, var_name_hdf5='Y', top=5, bottom=5, left=8, right=5)
fname_new = cm.save_memmap([Y], base_name=fname[:-4], order='C')

centers = [[ 79,  56],
    [119, 176],
    [ 91,  58],
    [122,  92],
    [111,  39],
    [ 52,  12],
    [ 99,   5],
    [ 11,  22],
    [117,  19],
    [ 85,  30],
    [176,  80],
    [ 31,  10],
    [115,  55],
    [ 30, 152],
    [  7,  89],
    [ 64,  76],
    [ 29,  33],
    [ 38, 113],
    [152,  17],
    [ 14, 138],
    [ 57,  68],
    [167, 113],
    [104, 110]]
params_dict = dict(
    fnames=[fname_new], method_init='corr_pnr', gSig=(4, 4), gSiz=(15, 15),
    p=1, tsub=1, ssub=1, only_init=True, nb=0, min_corr=0, min_pnr=0,
    normalize_init=False, ring_size_factor=2, center_psf=True, ssub_B=2,
    init_iter=1, del_duplicates=True, seed_method=centers)
opts = cnmf.params.CNMFParams(params_dict=params_dict)
cnm = cnmf.CNMF(n_processes=1, params=opts)
cnm.fit_file()

You need to set rf=None because patches aren't supported. And also only_init=True (to avoid detrimental temporal/spatial updates that don't use the accurate ring-model).

Just FYI, the other already previously supported values for seed_method are auto, manual and semi. The latter should be combined with init_iter=2 to perform one round of automatic initialization and a second round to manually add further components.