Tracing membrane lines in a membrane-marker image
The Membrane engine reads a grayscale image of the membrane marker, the Shades input. That image might be a channel produced by Color SeparationLoading... on an IHC sample. The engine tests a set of kernels whose widths span a range you set, and returns two binary masks of the membranes it finds (see The Two Outputs).
Objects can be filtered by area and by mean intensity. A border taken from neighboring fields of view keeps membranes continuous across tile edges.
The engine traces the thin lines of membrane staining and returns two pictures: the membranes as found, and a one-pixel-wide centerline of them.
The engine produces two binary masks of the same membranes. They differ in width, and each suits different downstream work.
| Output | StrataQuest label | What it contains |
|---|---|---|
| Membrane mask | Membrane thick | Every pixel the engine detected as membrane, at the width it was detected. |
| Membrane skeleton | Membrane thin | The membrane mask thinned to a centerline one pixel wide. It keeps the shape and connections of the membranes, but not their width. |
The Membrane MeasurementsLoading... engines take a membrane mask as input and report, for each event, how much of its surroundings the membrane covers, the membrane's mean intensity, the distances to it, and its mean thickness. Coverage and thickness depend on width, which only the full-width mask carries. The skeleton suits work on the membrane network itself: the Basic Operations skeleton operationsLoading... find its branch points and end points, and can extend spurs to close gaps.
You get two versions of the same membranes: the full-width mask, for measuring, and a one-pixel-wide centerline, for following how the membranes connect.
| Parameter | Description |
|---|---|
| Shades (membranes) | Grayscale image of the membrane marker. |
| Remove smaller than | Area threshold for detected objects. |
| Intensity threshold (Lower / Upper) | Limits on detected objects’ mean intensity. |
| Min variation (for ᴧ-like) | Minimum variation threshold for ᴧ-like membranes (default 0). |
| Kernel width (LL / UL) | Lower and upper limit of the kernel-width range. |
| Kernel height | Height of the kernel. |
| Standard deviation | Weight distribution of the Gaussian model used to search for membranes. |
| Border size / Border option | Pixels borrowed from neighboring fields of view, and the model used where there is no neighbor. |
Set Kernel width (LL / UL) to span the thickness of the membranes in your image. Use Intensity threshold and Remove smaller than to drop faint or tiny detections.
HER2 membrane scoring in breast cancer tissue, following the published ASCO/CAP convention:
Under the published HER2 scoring convention, only membrane staining is scored, so the membrane compartment is the measurement that counts here. Cytoplasmic HER2 signal is not part of the score and would confound it if it were included. The Membrane engine isolates the membrane signal that is scored.
Under the published HER2 scoring convention, staining is scored at the cell membrane, not in the cytoplasm. The Membrane engine separates that boundary signal, so each cell can be scored 0, 1+, 2+ or 3+ on its membrane alone.
