Textural analysis of lymphoid cells in serous effusions. A mathematical morphologic approach

Nuclear texture of reactive and well-differentiated neoplastic lymphocytes from serous effusions was studied by an approach based on principles of mathematical morphology. Density features were obtained before and after gray level nuclear image transformation by morphologic closing (dilation followed by erosion) and application of a top-hat function, which detects light or dense spots with a determined width and contrast. Five cases of benign reactive lymphocytic serous effusions and 11 cases of effusions in well-differentiated lymphocytic lymphomas were analyzed retrospectively. Each lymphoid cell was characterized by 24 densitometric features. Rank-order transformation was used for linear discriminant analysis given non-normal distributions of variables. In a first model formed by pooling all cells, discriminant function distinguished between 110 reactive and 216 malignant lymphocytes in the learning set and between 111 reactive and 226 malignant lymphocytes in the test set with better than 81% accuracy in both. In a second model, correct classification of cases as reactive or malignant was achieved in 5/5 reactive and 11/11 malignant lymphoid effusions. The results indicate that mathematical morphologic transformations of the gray level image may be an effective adjunct to other textural descriptors of cellular atypia, especially in the differential diagnosis of lymphoid serous effusions.