Importance of different nuclear morphologic patterns in grading prostatic adenocarcinoma. An expanded model for computer graphic filters.

The aim of this work was to continue the development of an interactive workstation for the nuclear grading of prostatic lesions by including a large range of nuclear patterns. A previous model was based on four groups: hyperplasia, Mostofi grade 1, Mostofi grade 2 and Mostofi grade 3. Each group included the most common nuclear patterns of the lesions. One set used to test the model included cases showing patterns different from the typical ones of the model. Poor results were obtained for low and medium grades. A review of all the cases in our database led to the conclusion that different nuclear patterns can belong to the same "nuclear grade." Thus, in this work the model was expanded to include six groups: hyperplasia, two subgroups for Mostofi grade 1, two subgroups for Mostofi grade 2 and Mostofi grade 3. A set of 900 nuclei, 150 in each group, was selected to test the model. An additional 300 nuclei, 50 in each group, were used for a test set. The overall success rate for classifying the nuclei in the test set using the new model was 93% as compared to a rate of 71% obtained for the similar test set, described above, using the previous model. Moreover, correlating karyometric features with nuclear morphology indicated a role for nucleoli in nuclear grading. The good results obtained with large and heterogeneous sets of cases indicate that the procedures used to develop this model may be adapted for the development of models for the nuclear grading of other tumors.