A texture analysis model for the classification of video images of B and T cells and the formation of subcategories

Abstract

B and T lymphocytes of known origin were used to test a pretopologic (mathematical) classification model that uses chromatin texture as a parameter. One data base was obtained by digitizing the cells with a standard TV camera coupled with a microscope, with the images transmitted to an image-analysis system (ASTI) coupled with a multi-20 computer. A second data base was obtained from the same slides using the TAS coupled with a PDP 11/34. The data bases were analyzed by the pretopologic program, which paramatizes the dispersion and aggregation of 15 possible optical densities (O.D.) and their relation to each other for each cell. Sub-categories were formed according to an algorithm that forms categories of images having the same pretopologic predominating sequences of O.D. values. In the ASTI system, 92% of the B-cell and 94% of the T-cell training samples were classified unambiguously; the test sample was classified as 60% B cells and 40% T cells, which is in close agreement with the spleen distribution as noted in the immunologic literature. The classification algorithm found 13 B-cell subcategories and 12 T-cell subcategories in the respective training sample; classification of the spleen test sample added no new subcategories. The TAS results were in agreement with the ASTI results within the limits of a restricted sample size. The overall B-cell and T-cell classification results were in agreement with those obtained by other authors . The subcategories found are unlikely to be fortuitous since only certain combinations of predominating O.D. values occurred and a large number of digitized nuclei had identical parameters. Nevertheless, the biologic significance of the subcategories cannot be assessed by mathematical methods alone, and these methods must be tested with appropriate biologic models.