printlogo
http://www.ethz.ch/index_EN
Institute of Microbiology
 
print
  

D - Basic concepts

Support Vector Machine, SVM. SVM is one of several machine learning algorithms. LibSVM is a popular implementation for support vector machine, is exclusively used by Enhanced CellClassifier and is firmly integrated in the program. Please check the webpage http://www.csie.ntu.edu.tw/~cjlin/libsvm for further useful information. We especially recommend the document "A practical guide to SVM classification" on this webpage, since CC basically follows the strategy explained in this paper.

Split data. CellProfiler saves data from the image analysis as a CellProfiler-output file. If one is doing many measurements or analyzes many images, these output-files easily become prohibitively large, causing Matlab-"Out of memory" errors even on powerful computers or at least significantly slow down further analysis. Therefore we decided to "split" the CellProfiler-output files. Thereby we convert the data and save them as one Matlab file for each CellProfiler cycle (=one file per image, or field of images). This "splitting" can be done from several places within Enhanced CellClassifier. Two kinds of files are generated:

The LabelMatrixImage = Segmented Image gives information about the positions of the objects to be classified. It is an image as big as the original image with the shapes of the identified objects. It is a two dimensional Matlab variable (matrix) and has zeros for each pixel which does not belong to an object and numbers for pixels at pixels belonging to an object, the value of the number corresponding to the object number. The LabelMatrixImage has to be made by CellProfiler with the ConvertToImage module (see preparations). The files for all LabelMatrixImage of one plate need to be in one folder.

Object: An object is defined by a CellProfiler pipeline. Objects might for instance be nuclei, cells or spots. Objects can be measured by CellProfiler and classified using Enhanced CellClassifier.

Model: A model is calculated by the SVM algorithm and contains the information how the object measurements can be translated to classification information. Models can be saved and loaded by Enhanced CellClassifier.

Vector: Vectors are important for summarizing and integrating data. A vector is binary (contains only the values zero or one). Vectors are calculated image wise and belong to a parent object. A vector contains as many elements as objects of this kind exist on a given image. The number of vectors that can be defined is not limited.

For example: On one image nuclei are identified. If there would be five nuclei on a given image, the vector for all nuclei would be 1 1 1 1 1. The user can now define a vector "mitotic_nuclei" for mitotic nuclei, using a classification algorithm. If nuclei 2 and 5 are recognized as mitotic, this vector for this image would be 0 1 0 0 1. The user might define a third vector, "non_mitotic", by subtracting the vector "mitotic_nuclei" from all nuclei. The vector "non_mitotic" for this image would be 1 0 1 1 0. Many information, for instance from the SVM-classifier or CellProfiler measurements, can be translated into vector information.

Image Variable: Image variables are important for summarizing and integrating data. Image Variables are defined by the user, the number of image variables is not limited. For each Image Variable one value is calculated per image (in contrast to a Vector which contains as many values as parent objects). Values can be calculated using Vectors or CellProfiler measurements.

Note: We use the term "Image Variable" even though an image might consist of several channels (DAPI, GFP, actin etc).

Well Variable: The concept of well variables is similar to that of an image variable, except data from a whole well are summarized. Well variables can be calculated from image variables or vectors, their number is not limited.

Plate Variable: Plate variables can only be calculated from well variables. For defining a plate variable, the user has to select a well variable, any number of wells and a way of summarizing (either median or summarizing).

 

Wichtiger Hinweis:
Diese Website wird in älteren Versionen von Netscape ohne graphische Elemente dargestellt. Die Funktionalität der Website ist aber trotzdem gewährleistet. Wenn Sie diese Website regelmässig benutzen, empfehlen wir Ihnen, auf Ihrem Computer einen aktuellen Browser zu installieren. Weitere Informationen finden Sie auf
folgender Seite.

Important Note:
The content in this site is accessible to any browser or Internet device, however, some graphics will display correctly only in the newer versions of Netscape. To get the most out of our site we suggest you upgrade to a newer browser.
More information

© 2014 Microbiology ETH Zürich | Imprint | Disclaimer | 23 December 2009
top