A first-pass mask from a single threshold is rarely the final form you want. Real images carry layered structure — variation across the field, isolated extreme pixels, regions whose shape is almost but not quite what the analysis is asking about. The mask becomes useful through a sequence of small transformations that each surface a particular property.
The typical pattern is: produce a starting mask, then shape it. Shape means morphology — dilate, erode, open, close, fill holesLoading... — to express the geometry you want. Combine — using AND, OR, XORLoading... — to keep only the intersection of two masks, or the symmetric difference. Constrain — cut narrow bridgesLoading... between regions the analysis should treat as separate objects. Then label, then measure.
The Operations Editor is built around this rhythm. A typical Detection slot is two engines, sometimes three, chained: detect, shape, label.