In microscopy images of tissue sections, we are trying to delineate overlapping semi-transparent objects of random sizes and shapes, a technically demanding task that even human annotators find challenging to do repeatably. Segmentation models that explicitly allow for overlap such as our developed Sequential UNet generates a potentially more realistic segmentation representation of the tissue. We show how this can improve the information one can extract from DNA specifically stained nuclei on a large lung cancer TMA (~800 spots). We will present how a one to many mapping could potentially improve marker evaluation from overlapping cells in spatial biology.