Telling one from another

Abstract

IT is still an open question that how images of individual cells can be identified automatically in an unsupervised manner. Furthermore, it is a great challenge to quantify cell morphology and intracellular structures either in three or even four dimensions given the great heterogeneity between cells and the heterogeneity in quality between images and samples. These important aspects of a quantitative and unbiased cell analysis for cytomics or cell systems analysis have been frequent issues in the recent past (1,2). Particularly, neurite morphology is very complex with a number of features relevant in information transduction and processing, such as the number of dendrites, their length, branching and branching points, number and type of synapses, or axon length (3). A method of quantitative analysis commonly used in neuronal studies to characterize the morphological characteristics of an imaged neuron—first used in 1953 to describe the differences in the visual and motor cortices of cats—involves initial quantification of a neuron. The analysis is performed by counting the number of dendrite intersections for concentric circles usually centered at the centroid of the cell body, of gradually increasing radius. This approach is now further elucidated and refined by Langhammer and coworkers (this issue, page 1160) using the Sholl analysis of arbor subregions of the dendrites, leading to new biological findings. Separating automatically or interactively close-by objects such as cells in confluent cell or tissue cultures can be a demanding task, even for an experienced observer, not to mention a digital one (4). Like a drop of water falling on a topographic relief flows along a path, such is the grey-shade of a microscopic image of cells, to finally reach a local minimum; the watershed of a relief corresponds to the limits of the adjacent minimum between imaged objects. These algorithms are termed Watershed algorithms and are in image analysis, applied to separate close-by objects. This can be applied for single images or three-dimensional image stacks (5). Now, Kachouie and coworkers (this issue, page 1148) propose a constrained watershed tool by which they can segment up to 99% of cells in confluent cultures, a method that also works for mosaic images. Du and coworkers (this issue, page 1137) use an alternative approach to split objects with ill-defined boundaries. This method is interactive and applies geodesic commute distance: a combination of the geodesic and the commute distance approach. This method outperforms presently applied segmentation methods identifying cell images and splitting them even if signals are weak and cells are overlapping. In his Communication to the Editor, Metze (this issue, page 1101) discusses some aspects of the recently published work by Wang et al. (6). This work focused on classification of follicular lesions of the thyroid. The authors describe a computerized method to detect and classify follicular adenoma of the thyroid, follicular carcinoma of the thyroid, and normal thyroid based on the nuclear chromatin distribution from digital images of tissue obtained by routine histological methods. In conclusion, the authors state that nuclear structure alone contains enough information to automatically classify the normal thyroid, follicular carcinoma, and follicular adenoma, as long as groups of nuclei (instead of individual ones) are used. Metze adds an interesting and important aspect to this work, namely that the lesion size is a critical and a potentially determinative aspect of correct lesion classification. Although the aforementioned articles have the automated recognition, segmentation, and classification of microscopic images on the micro (lm) scale as the major focus, two other studies address image manipulation or pattern recognition on the macro scale. Lahrmann and coworkers (this issue, page 1169) addressed the problem of correctly assigning tissue sample images from tissue micro arrays (TMAs) based on whole slide imaging data. TMAs have only been in use since not much more than 10 years. They are very valuable tools to test the binding of drugs to different rare biopsies in high-