Wavelets as chromatin texture descriptors for the automated identification of neoplastic nuclei.

Chromatin distribution reflects the organization of the DNA of a nucleus and contains important cellular diagnostic and prognostic information. Feulgen staining of breast tissue enables the chromatin distribution of the nucleus to be visualized in the form of texture. Describing texture in an objective and quantitative way by means of a set of texture parameters, combined with the study of the relationship of such parameters to the pathobiological cell properties, is useful both for reduction of the subjectivity inherently coupled to visual observation and for more accurate prognosis or diagnosis. We have presented an automated classification scheme for the diagnosis and grading of invasive breast cancer. The input to this scheme was a digitized microscopical image, from which nuclei were segmented. Chromatin texture was described using a set of textural parameters that include first- and second-order statistics of the image grey levels. The more recently developed wavelet energy parameters were also included in our study. Classification was performed by a Knn-classifier, which is a versatile multivariate statistical classification technique. We investigated the role of the tissue preparation technique and found that parameters derived from cytospins were better texture descriptors than those from sections. A 100% correct classification was achieved in a patient diagnosis experiment and 82% in a nuclear grading experiment.