Subclassification of follicular lymphomas by computerized image processing

Non-Hodgkin's lymphomas with a follicular pattern are subdivided on the basis of morphologic criteria into four subtypes. Each subtype has a distinctive natural history, response to therapy and survival, and precise histologic diagnosis is essential for optimal treatment. However, recent studies by a panel of seven expert hematopathologists showed that there was a 20% to 40% likelihood of disagreement in subtyping non-Hodgkin's follicular lymphomas; thus, the likelihood of a patient receiving inappropriate treatment is high. To resolve some of the problems relating to subjective morphologic diagnosis, we have continued to investigate the possibility of using computerized image analysis to perform automated subtyping of follicular lymphomas. In the portion of the study reported herein, a total of 37 cases were selected from the set subtyped by the panel of expert hematopathologists, and digitized images of slides from each case were processed by computer. The computer-generated subtypes were compared with those arrived at by the panel. For the cases analyzed, the computer subtyping was at least as good as, and in some instances superior to, that of most of the expert panelists.