Otsu's method, where it works well. The Engines Reference states the method plainly. Otsu's threshold splits pixels into foreground and background by minimizing the variance within each class, or equivalently maximizing the variance between them. It works well when the histogram has two peaks with a deep valley between them, as clearly positive staining on a clearly negative background often does.
Where it fails. The Otsu Threshold (global) chapter names the hard cases. When the positive area is small compared with the background, the histogram is no longer two-peaked, and when both classes are broad or noise fills the valley, the valley fades. Uneven illumination can add a third class. Narrowing Lower and Upper is the documented answer, and Background RemovalLoading... beforehand is another way to even out the background.
Per field or global. A threshold computed in each field of view follows local brightness, but neighboring fields can end up with different cutoffs. One global threshold treats every field alike, which is usually what you want when you compare stained area across fields or samples. The Total Area quick action "Use Otsu Threshold (Global)" adds that engine and points Use global variable at its output.
From events to a percentage. Total Area doesn't report a percentage. In general, percent positive area is the positive area divided by the analyzed area. In StrataQuest you'd take the first from Standard Measurements on the area events and the second from the region, and combine them, for example with the Statistics Report's arithmetic. For area within a region, the User Manual recommends the Crop events to ROI's shape strategy, so patches crossing the boundary count only inside it.
Otsu's method puts the cutoff in the valley between a bright peak and a dark peak in the histogram. It works poorly when the stained area is tiny or the lighting is uneven. To compare samples, use one global cutoff rather than one per field.
