QUICK SEARCH:   [advanced] Author: Keyword(s):
Home     Help     Feedback     Subscriptions     Archive     Search     Table of Contents

First published online October 24, 2007
doi: 10.1242/10.1242/jcs.013623

This Article Summary Full Text Full Text (PDF) Alert me when this article is cited Alert me if a correction is posted Services Email this article to a friend Similar articles in this journal Similar articles in PubMed Alert me to new issues of the journal Download to citation manager Citing Articles Citing Articles via Google Scholar Google Scholar Articles by Wollman, R. Articles by Stuurman, N. Search for Related Content PubMed PubMed Citation Articles by Wollman, R. Articles by Stuurman, N.

High throughput microscopy: from raw images to discoveries

¹ Department of Molecular and Cellular Biology, University of California, Davis, CA, USA
² The Howard Hughes Medical Institute and the Department of Cellular and Molecular Pharmacology, University of California, San Francisco, CA, USA

View larger version (16K): [in this window] [in a new window]   Fig. 1. Analysis pipeline. An example of the analysis pipeline, used in an RNAi screen of Drosophila spindle morphology. After acquisition, the images were streamlined to the image database. The analysis code accessed the raw images, segmented them to identify cells and classified the cells to distinguish mitotic cells from apoptotic and interphase cells. The mitotic cells were reorganized into galleries for visual analysis. The same cells from the galleries were analyzed further to identify specific mitotic phenotypes. Bootstrap statistics was used to identify wells that exhibited statistically significant extreme phenotypes.

View larger version (52K): [in this window] [in a new window]   Fig. 2. Segmentation. (A-D) Segmentation using the simple threshold method. The DAPI channel (B) from the cell shown in A is used for segmentation. The histogram (D) of intensity values for the DAPI channel shows bimodality. A background population of pixels averages an intensity of 20 and a foreground population of pixels averages an intensity of 100. The optimal threshold (an intensity of 50) is the intensity that separates those two population while maximizing the between-population variance and minimizing the within-population variance. (C) Binary image created by using the optimal threshold. (E-L) Segmentation using edge detection. Yeast cells imaged in DIC that are unsuitable for threshold-based segmentation can be segmented by the edge-detection algorithm. (E) Raw image of the yeast cells. Results of an edge detection (F) operation are then improved by morphological dilation (G) and a morphological operation that fills the `holes' in the image. Owing to the initial dilation, the binary masks are too large and are reduced by morphological erosion (I). The results of the segmentation are three unconnected yeast cells that can be labeled individually. Images K and L show the final edge and an overlay of the edge on the original cells. (M-P) Segmentation using watershed operation. Nuclei of Drosophila S2 cells stained with DAPI (M) are in close proximity and a simple threshold (N) does not segment them properly. A distance transformation assigns each pixel inside the cell a value equal to its distance from the edge. This creates an image that can be interpreted as 3D topographical landscape containing two `hills' (O). The watershed transformation separates the hill by a mathematical operation that is equivalent to `raising the water level' until all hills are separated; the result is seen in P.

View larger version (54K): [in this window] [in a new window]   Fig. 3. Cell classification. Cells were labeled with an antibody binding phosphorylated histone 3 (PH3, marking mitotic cells), an antibody binding tubulin and with DAPI (labeling DNA). (A-E) Scatter plot of a population of cells (A) showing three distinct populations: interphase cells pH3 intensity<25 and tubulin nuclear intensity 0–150 (see C); apoptotic cells, pH3 intensity 0-150 and tubulin intensity<20 (see D); and mitotic cells, high pH3 intensity and tubulin staining (see E). Cells can be classified as mitotic and non-mitotic using a simple artificial neural network. The result of the training is shown as a heat map (B) of the probability that a cell is mitotic. Mitotic (and apoptotic) cells can be discriminated from interphase cells solely on the basis of pH3 intensity; however, since both mitotic and apoptotic cells stain well with this antibody, an additional feature (tubulin staining) is needed to discriminate between the two. Notice that, although in this simple example a neural network is not strictly necessary and good classification could be achieved using two thresholds, training of artificial classifiers scales better with dimensionality of the inputs and does not require manual tinkering to determine classification parameters.