Projection Images

Projection Images#

Suite2p computes several reference images during registration and detection. These images are used for quality assessment and as inputs to Cellpose anatomical segmentation.

Processing Pipeline#

Images stored in ops.npy are computed at different stages:

Step

Description

Disable

Temporal Binning

Average consecutive frames to reduce noise

ops["nbinned"] = nframes

Temporal High-Pass

Remove slow baseline drift from movie

ops["high_pass"] = 0*

* The MBO fork allows disabling high_pass; standard Suite2p always applies it.

What’s stored:

  • meanImg - mean of binned movie (computed before HP filter)

  • meanImgE - enhanced mean from registration (median filter based)

  • max_proj - max of binned + HP filtered movie

  • Vcorr - correlation map or cellpose input image

Anatomical Detection Modes#

Mode

Image

Description

anatomical_only=1

log(max_proj / mean_img)

Activity relative to baseline

anatomical_only=2

mean_img

Temporal mean

anatomical_only=3

meanImgE

Enhanced mean (recommended)

anatomical_only=4

max_proj

Maximum projection
Anatomical Modes

Comparison of anatomical detection modes. mode=1: log ratio highlights active regions. mode=2: temporal mean. mode=3: enhanced mean with median filter. mode=4: maximum projection of HP-filtered movie.#

# mode 1: activity relative to baseline
img = np.log(np.maximum(1e-3, max_proj / np.maximum(1e-3, mean_img)))

# mode 2: temporal mean (before HP filter)
img = mean_img

# mode 3: enhanced mean with local median subtraction
from scipy.ndimage import median_filter
Imed = median_filter(mean_img, size=d)
img = (mean_img - Imed) / median_filter(np.abs(mean_img - Imed), size=d)

# mode 4: max of HP-filtered movie
img = max_proj

Spatial High-Pass Filter#

The spatial_hp_cp parameter applies additional high-pass filtering before Cellpose segmentation. This is separate from the temporal high_pass used during detection.

Spatial HP Filter

Effect of spatial_hp_cp values on the max projection. Higher values sharpen cell boundaries but may amplify noise.#

from scipy.ndimage import gaussian_filter

def apply_hp_filter(img, diameter, spatial_hp_cp):
    sigma = diameter * spatial_hp_cp
    return img - gaussian_filter(img, sigma)

Recommendations#

ops = {
    "anatomical_only": 3,      # use enhanced mean
    "spatial_hp_cp": 0,        # usually not needed with meanImgE
    "diameter": 6,             # cell size in pixels
    "cellprob_threshold": 0.0,
    "flow_threshold": 0.4,
}

If detection is poor:

  • Try anatomical_only=4 (max projection)

  • Increase spatial_hp_cp to 1-3

  • Adjust diameter to match cell sizes

  • Try anatomical_only=1 for data with strong baseline fluorescence

See Also#

Contents