Two Approaches for Automatic Nuclei Cell Counting in Low Resolution Fluorescence Images

This paper deals with the problem of counting nuclei in low-resolution fluorescence images. The low resolution is a consequence of the usage of a large field of view, this to reduce the number of experiments, implying time saving and money saving. But the small size of nuclei increases the risk of error in counting. In this work we used some image processing basic operators in order to extract potential shapes. These shape are then characterized (size, shape, edges) to decide if they are noise or not. Two approaches are presented for this: the first one correspond to a translation of some basic knowledge used by practitioners; the second one uses a supervised classification technique where it is the computer who discovers the knowledge by the analysis of a database of cases. This leads to a fast technique which gives very good results, validated by cell biology experts