CT image of the right ventricle

First:Test image. It is an ultra-fast CT image of the right ventricle of the canine.
Second: Gradient image.
Third: Interactively imposed markers (really dark pixel areas).
Fourth: Meyer's watershed transform. The water coming from the markers floods all the irrelevant catchment basins except the highest-crest line around the object marker.  The segmentation result looks visually sensible.

Questions:

1. How can the markers be automatically detected?
2. What could be the final wateshed lines if the object marker is a circle instead of an elongated bar?