Multiresolution Texture Analysis of four classes of Mice liver cells using different cell cluster representations

We have analyzed first and second order multiresolution texture features in order to discriminate normal and pathological liver cell nuclei. Several combinations of first and second order texture features discriminate animals from the four different groups. The best feature pair (graylevel Variance of the euchromatine and Diagonal moment of the GLCM) gave a correct classification result of 95 %. Representing the cell clusters using different representation methods seems to improve the classification result. Classification with one feature measured at two different resolutions gives the same result as two different features measured at the same resolution. Because of the small dataset further testing is needed to confirm the results.