Detecting the transition stage of cells and cell parts by prototype-based classification

Abstract

Unsupervised classification is the choice when knowledge about the class numbers and the class properties is missing. However, using clustering might not lead to the correct class and needs interacting with the domain experts to figure out the classes that make sense for the respective domain. We propose to use a prototype-based learning and classification method in order to figure out the right number of classes and the class description. An expert might start with picking out a prototypical image or object for the class he is expecting. Later on, he might pick out some more prototypes that might represent the variance of the class. By doing so might be incrementally learnt the class border and the knowledge about the class. It does not need the expert so heavy interaction with the system. Such a method is especially useful when the domain has very noisy objects and images. We present in the paper the method for prototype-based classification, the methodology, and describe the success of the method on a biological application - the detection of different dynamic signatures of mitochondrial movement. © 2014 Springer International Publishing Switzerland.