Nuclear texture in poorly differentiated small round cell tumors. Image analysis study of fine needle aspiration...

Gradient analysis and pattern spectrum decomposition based on mathematical morphology concepts were used to explore nuclear texture patterns in a pool of 108 cells obtained by fine needle aspiration of five undifferentiated small round cell tumors of childhood, including one case each of Wilms' tumor, neuroblastoma, lymphoblastic lymphoma, Ewing's sarcoma and rhabdomyosarcoma. The aim of the study was to determine the presumptive value of nuclear pattern to correctly allocate each isolated cell to each of the five patients. The cells were examples of five histogenetically different tumors, all undifferentiated and with a close microscopic resemblance to one another. High gradient structures (heterochromatin-euchromatin and nuclear membrane edges) were estimated by a difference-of-boxes filter, and pattern spectrum decomposition was obtained by successive openings and closings performed on the input gray tone image. One important feature of these procedures was that no prior selection by thresholding of the structures to be studied was required, thus obviating subjective bias. Percentages of correctly allocated cells by canonical analysis ranged from 70.0% (rhabdomyosarcoma) to 92.9% (Ewing's sarcoma). Although the five cases could be distinguished using seven texture variables, this does not imply generalization of the results for the differential diagnosis of these tumors. Nonetheless, the possibility that undifferentiated small round cells present distinctive nuclear patterns when studied by sensitive image analysis techniques is suggested by our results.