Confocal fluorescence microscopy of leaf cells: An application of Bayesian image analysis

Abstract

Confocal fluorescence microscopy is a recent and important imaging tool for visualizing three-dimensional specimens without the need for physical sectioning, so that changes in living cells can be studied over time. The application of interest here is a study of the changes in the stomatal guard cells of plant leaves during their opening and closing response sequences. Quantitative estimates of the size and shape of these cells rely on accurate classification (or segmentation) of the images into areas which are parts of cells and areas which are background. This segmentation is complicated in confocal microscopy because the images appear to be 'smudged' or 'dirty'; this degradation is due largely to diffraction and attenuation of the recorded signal caused by the specimen itself. Correcting for this degradation is difficult without knowing the specimen-dependent parameters involved in the degradation process. A fully Bayesian approach is proposed for tackling this problem of blind deconvolution, i.e. of concurrently estimating the degradation parameters while segmenting two-dimensional sections. The end-products are interval estimates of size and shape which acknowledge some of the uncertainty involved in the segmentation. The results are promising, generating credible intervals which are sufficiently narrow to be useful in practice.