Applying watershed algorithms to the segmentation of clustered nuclei

Abstract

Cluster division is a critical issue in fluor escence micr oscopy-based analytical cytology when pr eparation pr otocols do not pr ovide appr opriate separation of objects. Overlooking cluster ed nuclei and analyzing only isolated nuclei may dramatically incr ease analysis time or af fect the statistical validation of the r esults. Automatic segmentation of cluster ed nuclei r equir es the implementation of specific image segmentation tools. Most algorithms ar e inspir ed by one of the two following strategies: 1) cluster division by the detection of inter nuclei gradients; or 2) division by definition of domains of influence (geometrical appr oach). Both strategies lead to completely dif fer ent implementations, and usually algorithms based on a single view strategy fail to corr ectly segment most cluster ed nuclei, or per for m well just for a specific type of sample. An algorithm based on morphological watersheds has been implemented and tested on the segmentation of micr oscopic nuclei clusters. This algorithm pr ovides a tool that can be used for the implementation of both gradient- and domain-based algorithms, and, mor e importantly, for the implementation of mixed (gradient- and shape-based) algorithms. Using this algorithm, almost 90% of the test clusters wer e corr ectly segmented in peripheral blood and bone marr ow pr eparations. The algorithm was valid for both types of samples, using the appr opriate markers and transfor mations.