Computerized classification of nuclear profiles in non-Hodgkin's lymphomas

The authors present experimental technics for the diagnosis of non-Hodgkin's lymphomas, based on instrumental classification of nuclear profiles using a video-based system for computerized interactive morphometry (CIM). In their system, the real time video image of a specimen is superimposed to a graphics overlay generated by a computer, consisting of a test area with four visual markers for random sampling of cells and a menu with several options to send direct commands to the system. Using a touch-sensitive screen mounted on a video monitor as an interactive peripheral, a trained observer traces 100 randomly selected lymphoid cells, counts mitoses in 25 microscopic fields, and categorizes the lesion as diffuse or nodular. Each cell is instrumentally classified into either small cell noncleaved, small cell cleaved, or large, based on the length of their nuclear profiles, their enclosed nuclear area, and a circularity factor. Thereafter the computer provides a 'diagnosis', based on hierarchic analysis of the data. The morphometric data are also interpreted by alternate statistical methods of discriminatory classificatory analysis that provide a diagnosis and a probability statement derived from matching unknown cases with a data base. Forty-two lymphoid lesions have been categorized with the CIM system. Studies of interobserver and intraobserver variations in data collection are discussed. The potential advantages of CIM for the objective classification of non-Hodgkin's lymphomas are discussed.