ADVANCES IN 3-D NUCLEAR SEGMENTATION AND CATFISH ANALYSIS

G. Lin¹; K. Olson²; M.K. Chawla²; S.N. Burke²; V.L. Sutherland²; B.L. McNaughton²; P.F. Worley³; J.F. Guzowski⁴; B. Roysam¹*; C.A. Barnes²

1. ECSE, Rensselaer Polytechnic Institute, Troy, NY, USA
2. NSMA , Univ. Arizona, Tucson, AZ, USA
3. Neurosci & Neurol, Johns Hopkins Univ, Baltimore, MD, USA
4. Neurosci, Univ New Mexico, Albuquerque, NM, USA

Automated 3-D cellular compartmental analysis of temporal activity by fluorescence in situ hybridization (3-D catFISH) is a tool for studying differential activation with cellular resolution. This system can be used for neuronal tract tracing, 3D visualization, automated region-of-interest detection, comprehensive signal association and quantification across the different image channels. We report methods that improve the nuclear segmentation accuracy, expand the number of fluorphores, and enable automated batch processing of large numbers of confocal image stacks. Improved segmentation accuracy was attained by a novel algorithm for analyzing connected clusters of nuclei containing multiple cell types. This method integrates the cell classification and fragment merging steps. Correcting for the image anisotropy by image resampling, and use of novel object features further improved the accuracy. Finally, the software was extended to handle a potentially unlimited number of FISH channels. To achieve high-throughput processing batch, a training image is drawn from each batch, and segmented. Object models are constructed, and the FISH threshold is set to match the consensus of multiple expert observers. After training, these data (object models and parameters) are applied to other images during automated batch processing. A novel display annotation was developed to enable rapid inspection and editing of the results. A detailed validation study on a series of 3-D image stacks taken from CA1, CA3 region of rat hippocampus and gustatory cortex regions demonstrated a nuclear segmentation accuracy of 94%, and FISH classification accuracy of 96% when compared to a multi-observer consensus.

Support Contributed By: AG023309 & AG009219

Key words: hippocampus, automated, cellular, fluorescence