Skip to main content

10  Textural Features

FLIM Playground extracts single-cell texture features from each channel’s intensity image and ROI mask. For an example application, see this study using CellProfiler.

For QPI channels, use the separate Spatial texture extractor, which measures the corrected OPD-derived mass distribution.

The first step is to create a single-cell intensity image for each cell based on the ROI mask and the intensity image. Then a list of texture features (the list may be extended in the future) is calculated:

10.1 Granularity

\(Granularity_n\) is the percentage of intensity contributed by bright objects of \(n\) pixels in radius. It is computed by performing a morphological opening with a disk of radius n on the cell intensity image and measuring the intensity lost as a percentage of the total intensity of the original cell ROI.

\(n = 1, 3, 5, 7\) are computed.

Higher granularity at smaller scales implies greater fragmentation.

10.2 Radial Distribution

Each cell intensity image is divided into \(4\) concentric rings, and each ring’s fraction of the total intensity is calculated. A higher fraction in ring 1 indicates more signal at the cell center; a higher fraction in ring 4 indicates more at the edge.

10.3 Mass Displacement

Mass displacement is the Euclidean distance (in pixels) between the geometric and intensity-weighted centroids, both calculated using pixels with intensity greater than zero within the cell ROI. If the ROI has no such pixels, the result is a missing value (NaN).

10.4 Intensity Sum

Intensity sum is defined as the sum of pixel intensities within the cell intensity image.

Note

For a 2D decay channel there is no intensity image or ROI mask, so Granularity, Radial Distribution, and Mass Displacement cannot be computed. Selecting Intensity Texture for such a channel therefore yields only intensity_sum, defined here as the sum of the cell’s decay curve (its total photon count across all time bins). The output column is Intensity texture_{channel}: intensity_sum.

In the final dataset, feature names are prefixed by the combination of the feature extractor name (i.e. Intensity texture) and the channel name, allowing Data Analysis to group the features. For example, Intensity texture_nadh: granularity_1 means the percentage of intensity removed when bright objects of 1-pixel radius are removed in the NAD(P)H channel.

This post provides a visual explanation of and code to calculate some of the texture features above.