Subclassification of follicular lymphomas by computerized microscopy.

The pilot feasibility study reported was carried out to determine whether computerized morphometry could be used to discriminate between different subtypes of follicular lymphomas. Thirteen cases of follicular lymphomas of the small cell and large cell types were examined by map-guided computerized microscopy. Initially, each specimen was digitized over a 5,120- X -5,120-microns area. Follicle maps depicting well-defined follicles were generated from these digitized images and used to guide a robot microscope to examine follicular center cells using from 20 to 100 fields, each 160 X 160 microns. Mathematical morphology was used to estimate the number of connected regions within each field examined. Based on the assumption that each connected region corresponded to a cell nucleus, it was expected that the number of connected regions found per unit area would correlate with the number of cells per unit area. Three prototypical small cell lymphomas and three prototypical large cell lymphomas were used as a training set to calibrate the methodology. Data obtained on these prototypical cases were used to establish statistical decision boundaries. Seven test cases, judged to be less typical but classifiable by experts, were then examined. Five of the seven were classified correctly, and the remaining two were identified as outliers. These results suggest that subclassification of follicular lymphomas may be aided by computerized microscopy. This pilot feasibility study is currently being expanded using a much larger sample of case material.