Haematopoietic cell clusters quantification using image analysis

Automatising the analysis of haematopoietic cells culture is a step which is necessary in order to develop the use of this kind of measurement in toxicological assessment. The purpose is the classification of cell aggregates into three groups namely micro-clusters, macro-clusters and colonies. This is classically done using human vision. However, reproducibility is not good and comparisons between experts or laboratories remains very difficult. In this work, we propose a method based on learning machine: different parameters representative of these three kinds of clusters are tested in order to choose the best discriminative ones, then a learning database is created in order to parametrise the software, before using it on actual clusters images. The algorithm used is the support vector machine (SVM). Results show an excellent ability because almost 90% of the test database is well classified using three geometric parameters namely area, minimum and maximum centroid distances.