Automatic Cellular Aggregates Quantification for Toxicology using Statistical Learning

Quantification of haematopoietic clusters is largely used in toxicology. However, visually counting and differentiating aggregates is a very tedious and subjective activity because of the difficulties to evaluate the limits between different types of cell clusters. Proposed here, is an automatic solution with a digital imaging system based on the use of statistical learning techniques. We evaluate the performances of several statistical classifiers (SVMs) with an emphasis on the definition of relevant cluster-related features. Performance demonstration is carried out over a reference test set of several tens of cluster images. System efficiency speaks favorably of the ability of the current approach to routine work