Top-hat is most useful in fluorescence imaging when the field of view has a brightness gradient: illumination variation, broad autofluorescence, or any background that varies on a much larger scale than the cells you want to detect. The cells are bright spots, and the background is the slowly varying canvas beneath them. Top-hat separates the two.
Use a structuring element a few cell-diameters wide. The cells, small and bright, survive the subtraction; the gradient, large and slow, ends up in the opening and is removed. Thresholding the original would need a different value in every region, because its background varies. The top-hat result has a flat baseline, so a single thresholdLoading... works across the whole field.
The same logic holds in brightfield imaging, where small dark features such as dots or granules sit on a varying canvas. There the mirror image of top-hat, a closing minus the original, separates them the same way; in the BOM, build it from Close (dilate, erode) followed by Subtract images.