Premium
Hierarchical, model‐based merging of multiple fragments for improved three‐dimensional segmentation of nuclei
Author(s) -
Lin Gang,
Chawla Monica K.,
Olson Kathy,
Guzowski John F.,
Barnes Carol A.,
Roysam Badrinath
Publication year - 2005
Publication title -
cytometry part a
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 1.316
H-Index - 90
eISSN - 1552-4930
pISSN - 1552-4922
DOI - 10.1002/cyto.a.20099
Subject(s) - computer science , segmentation , pattern recognition (psychology) , artificial intelligence , computation , tree (set theory) , cluster analysis , algorithm , mathematics , mathematical analysis
Background Automated segmentation of fluorescently labeled cell nuclei in three‐dimensional confocal images is essential for numerous studies, e.g., spatiotemporal fluorescence in situ hybridization quantification of immediate early gene transcription. High accuracy and automation levels are required in high‐throughput and large‐scale studies. Common sources of segmentation error include tight clustering and fragmentation of nuclei. Previous region‐based methods are limited because they perform merging of two nuclear fragments at a time. To achieve higher accuracy without sacrificing scale, more sophisticated yet computationally efficient algorithms are needed. Methods A recursive tree‐based algorithm that can consider multiple object fragments simultaneously is described. Starting with oversegmented data, it searches efficiently for the optimal merging pattern guided by a quantitative scoring criterion based on object modeling. Computation is bounded by limiting the depth of the merging tree. Results The proposed method was found to perform consistently better, achieving merging accuracy in the range of 92% to 100% compared with our previous algorithm, which varied in the range of 75% to 97%, even with a modest merging tree depth of 3. The overall average accuracy improved from 90% to 96%, with roughly the same computational cost for a set of representative images drawn from the CA1, CA3, and parietal cortex regions of the rat hippocampus. Conclusion Hierarchical tree model‐based algorithms significantly improve the accuracy of automated nuclear segmentation without sacrificing speed. © 2004 Wiley‐Liss, Inc.