Pick the operation by the property that separates the events you want from the rest. Each is named for what it removes, so read the name as its opposite: Remove labels (smaller) keeps the larger events.
Surviving events should be at least a certain size: Remove labels (smaller), with Area threshold (px).
Surviving events should be no larger than a certain size: Remove labels (larger).
Surviving events should be at least a certain brightness: Remove labels (weaker), with Intensity threshold and a gray image as Image 2.
Surviving events should be no brighter than a certain level: Remove labels (stronger).
Surviving events should overlap a known feature mask: Remove objects (without seeds), with the feature mask as Image 2 (seeds). ("Drop events that don't contain a seed" means "keep events that do.") One shared pixel is enough to count as a seed.
Surviving events should NOT overlap a known feature mask: Remove objects (with seeds).
Compound criteria become chains. "Cells larger than N pixels and brighter than X and overlapping the tumor mask" is three operations in sequence, one per criterion. Area is in pixels, so convert a µm² target with the sample area per pixelLoading... first.
Two criteria aren't in this group. Edge contact is handled by Remove objects touching border, and a range on any other measurement by Remove objects v2. Both are engines in the layer's Post-Processing section; see Remove ObjectsLoading....
In image analysis generally, keeping whole objects instead of trimming pixels leaves every survivor exactly as detected, so its area and intensity are unchanged. Keeping the objects a marker touches is what mathematical morphology calls reconstruction from markers; the seeded pair are that idea in one step. Survivors keep their original labels, and the reference names Remove labels as a source of missing labels, so run Re-Label (missing)Loading... afterward if anything downstream requires consecutive numbers.
Source: StrataQuest 8.1 Engines Reference, ch. 6.9 (pp. 284–303); ch. 6.6 (p. 198); ch. 37–38 (pp. 504–507).