Fig. 3: Examples of how the cluster algorithm attempts to group similar objects within a segment.
From: The in vitro micronucleus assay using imaging flow cytometry and deep learning
Beginning with 1500 unassigned objects (a), the Cluster algorithm groups objects of similar morphologies into clusters of similar objects located close to one another on the object map. By assigning 250 (b), 750 (c), and 1250 (d) objects to the appropriate ground truth model classes, the accuracy with which the unassigned objects are grouped by the algorithm increases. In panels (c) and (d), distinct islands have become more visible and it can be seen that MONO, BN and POLY cells are well-separated from one another, while objects with irregular morphology have also been placed into their own discrete island.
