Selection and Quantification of Objects in Microscopic Images: from Multi-Criteria to Multi-Threshold Analysis

Due to the increased number of applications of both microscopic imaging and image analysis including biomedical studies, the design of specialized algorithms and tools to facilitate quantitative assessment of objects in the image content is of urgent need. Recently, a number of approaches ranging from object counting by machine learning methods to statistical image analysis have been suggested and successfully implemented to resolve the cell quantification problem. Here, we revisit the above problem considering samples where objects presented in the same images have to be explicitly distinguished and quantified without involving any dedicated experimental setting like differential fluorescent staining. We consider several possible classification criteria and show explicitly how their combination in a single algorithm can be used to improve results in complex images where single criteria-based rules inevitably fail. Finally, we suggest a possible approach for the analysis of non-homogeneous images based on combining object selection results for different threshold values thus enhancing the algorithm from multi-criteria to multi-threshold analysis. To demonstrate the performance of the suggested solutions, we show several prominent examples of complex structures ranging from images containing both live and apoptotic cells as well as containing mixtures of globular and fibrous forms of heat-shock protein IbpA.