Glossary

Glossary#

skewness #: How often do you get an outlier in the pixel activity across time, good for calcium indicators because theyre flat relative to a few events over the course of a recording.
mean_img#: The average of the registered movie over time. Used as a baseline representation of the structural signal.
max_proj#: The maximum value per pixel over time. Highlights regions with high fluorescence, useful for detecting active areas.
meanImgE#: Enhanced mean image generated by Suite2p using filtering and sharpening operations. Useful for anatomical segmentation when available.
anatomical segmentation#: ROI detection based solely on static images (e.g. mean or max projection), ignoring fluorescence time dynamics.
functional segmentation#: ROI detection based on pixelwise correlations across time, identifying spatially contiguous regions with similar activity patterns.
seed#: A pixel or region in the image used as the starting point for growing a potential ROI. Seeding is based on high-intensity regions in the filtered image, and forms the initial guess for the shape and location of a neuron.
Delta F over F0#: \(\Delta F / F_0\) (Delta F over F naught) refers to the fractional change in fluorescence relative to a baseline signal F₀. This is a common normalization used in calcium imaging and neural activity recordings.
segmentation#: The general process of dividing an image based on the contents of that image, in our case, based on neuron location.
source-extraction#: Umbrella term for all of the individual processes that produce a segmented image.
deconvolution#: The process performed after segmentation to the resulting traces to infer spike times from flourescence values.
Rigid-registration#: The object retains shape and size.
Non-rigid-registration#: The object is moved and transforms shape or size.
pixel-resolution#: The length of each pixel, in micron (px/um).
roi_stats#: Dictionary containing per-ROI statistics computed during detection, including area, compactness, skewness, and spatial coordinates.
logistic regression#: A statistical model that predicts binary outcomes (cell vs non-cell) by fitting a sigmoid function to input features. Suite2p’s classifier uses logistic regression on ROI morphological features. The model outputs a probability between 0-1, thresholded at 0.5 by default to make the cell/non-cell decision. Training involves fitting the sigmoid curve to labeled examples so that cells map to high probabilities and non-cells map to low probabilities.
compact#: Spatial compactness of an ROI, measuring how circular/concentrated the pixels are. Computed as (4 * pi * area) / perimeter^2. A perfect circle has compactness of 1.0; elongated or fragmented shapes have lower values. Used by the classifier because real neurons tend to be compact while artifacts are often irregular.
npix_norm#: Normalized pixel count of an ROI. The raw pixel count divided by expected size based on the diameter parameter. Helps the classifier reject ROIs that are abnormally small (noise) or large (merged cells, blood vessels).