Virtual Channel

Suppose you have a nuclear stain in one channel and a membrane stain in another, and you want one image that shows both. Virtual Channel builds it. Add it in the layer's Pre-Processing section (Add, Operations (basic), Virtual Channel).

The engine has one parameter, Select gray images: the list of grayscale imagesLoading... available in the current layer. Those can be original fluorescence channelsLoading... or outputs of other engines. Selected images must have the same type, and if you select only one, the engine outputs it without any processing. Images generated in other layers must be imported into the current layer before they can be selected.

The output is a virtual channel combining the information of the selected images. In the reference example, a nuclear marker channel and a membrane marker channel go in, and one image showing both comes out. The Engines Reference doesn't say how the values are combined, whether by sum, maximum or something else, or how values beyond the bit depth are handled. If you need to know, compare output and input values in the Pixel Inspector.

The name has a second, looser use in the reference. Elsewhere it calls any image an engine generates a virtual channel, such as the Ki67 image from Color SeparationLoading..., and notes that such images are visible only in the layer that made them. When documentation says "virtual channel", check whether it means this engine.

Source: StrataQuest 8.1 Engines Reference, ch. 42 (pp. 544–546), ch. 23 (p. 437), ch. 41 (pp. 538, 542), ch. 14 (p. 405).

Ways to combine two images. In image processing generally, there are a few standard ways to merge channels pixel by pixel. A sum adds the signals, and can overflow the range of the bit depth unless it is scaled. A mean keeps the range but halves a signal present in only one channel. A pixel-wise maximum keeps the brighter of the two at every pixel, so each structure keeps its own intensity.

Where StrataQuest documents each. Because Virtual Channel doesn't state its rule, use the Basic Operations Module when the rule matters. Add images sums two images, Add weighted images sums them with a weight for each, and Max (every pixel)Loading... keeps the brighter value. Subtract images and Divide images cover the other directions. Each states its inputs and output, so the result is predictable.

Why merge at all. In general, a combined image helps when the structure you want to find is defined by two stains together, such as a cell outlined by membrane with a nucleus inside, and the next engine takes a single input. It can also hurt. Merging a bright channel with a faint one can bury the faint signal, and whatever background one channel carries, the merged image inherits.

When division is dangerous. In ratio imaging generally, dividing by values near zero produces huge, meaningless numbers. A small constant added to the denominator prevents that while barely changing the ratio where both signals are strong. That is the purpose of the Add epsilon to divisor option of Divide images, which is on by default.

Here is how a combined lymphocyte-marker image could be made and used.

1. Start from CD3, CD8 and CD20 as separate grayscale channels of the same type.
2. Add Virtual Channel and select the three images. Check in the Pixel Inspector how the combined values relate to the inputs.
3. Use the combined image as the input of Total AreaLoading... to measure the area occupied by any of the three markers.
4. To map where immune cells crowd together, detect them as events and use those events as the Coded input of Density of EventsLoading.... That engine takes events, not a gray image.

A tumor-to-immune ratio image needs division and a constant, which Virtual Channel doesn't document. Build it with Divide images in the ArithmeticLoading... group of the Basic Operations Module, with Add epsilon to divisor on. A ratio per cell, rather than per pixel, is a job for Derived MeasurementsLoading....