Computer Classification Of Rosette Forming Cells Using Microscope Images

This paper presents a computer classification system of rosette forming cells. In clinical immunology, it is important to distinguish lymphocytes into some subpopulations. For this purpose, rosette formation test is one of the most popular and useful methods. But recognizing rosetted and non-rosetted process depends on microscopic examination. We notice that an accumulative histgram of gray-levels presents characteristics of each rosette well. The system proposed here extracts rosettes from microscope image, constructs an accumulative histgrams of gray-levels for each rosette, and classifies rosettes into five groups using several features derived from the accumulative histgrams. Touching rosettes are also treated. The system is tested using 322 samples from 10 images. Total accuracy of classification is 73.3 %. And total correctrate of rosetted and non-rosetted dicision is 86.0 %. It is confirmed that computer classification of rosette forming cells is possible by the proposed method.