WND-CHARM features and OMERO workflow
WND-CHARM - an open-source image classification tool
WND-CHARM is a generalized pattern recognition system for images developed by the Goldberg group at the NIH/NIA. The emphasis of WND-CHARM is on generality: it has been used to analyze images from fluorescence microscopy, phase-contrast, DIC, H&E-stained histological sections, as well as other image modalities such as X-rays, CT scans, even modern art.
In contrast to traditional image processing software, analysis is performed by training a classifier with example images or control images from an image-based assay. In each case, the parameters required are limited to the arrangement of images into groups (classes) to indicate expected variation within the group as well as the differences that are considered relevant to the imaging problem. The grouping of images into classes is sufficient for the analysis to proceed automatically from computing of low-level image features, to selection of image features based on their discriminative power, to training a classifier for scoring subsequent images not previously used in training. The primary output of a WND-CHARM classifier is the probability of a test image’s membership in each of the pre-defined groups/classes.
The primary parameters of WND-CHARM coincide with one of the primary functions of OMERO: organization of images into groups relevant to the experimentalist. The ultimate goal is to use manual image annotations and organizational structure as stored in OMERO as a bootstrap to train classifiers that would automatically annotate images as they are imported into the system.
WND-CHARM is developed by the Goldberg Lab/Image Informatics and Computational Biology Unit at the National Institute on Aging, Baltimore USA.
WND-CHARM stands for Weighted Neighbor Distance using Compound Hierarchy of Algorithms Representing Morphology