Most BOM detection and measurement engines specifically expect grayscale on input — a single number per pixel. The reason is operational: the engine's question is usually how bright, not what color, and a single-channel input lets the engine reason about intensity without untangling channel correlations.
So a typical fluorescence pipeline starts with channel extraction — pulling, say, the DAPI or FITC channel from a multi-channel acquisition — and feeding that single grayscale image into Pre-Processing filters and Detection engines. A typical brightfield pipeline starts with color-space conversionLoading... to grayscale, or with stain separation that produces a grayscale image per chromogen.
The exception is engines explicitly designed for color, like the color conversionsLoading... or the ClassifierLoading... when trained on RGB features. Those want the multi-channel input. Most don't.