The DNA histogram obtained by flow cytometry can be considered as the product of an "ideal" measurement column vector and a measurement distortion matrix. In order to extract the ideal histogram from the real data, the measurement distortion matrix is commonly presumed to be a family of Gaussian coefficients that are centered on the diagonal. We have designed a feedback-controlled curve-fitting procedure that reconstructs the ideal histogram from the real data through successive iterations. The optimum coefficient of variation (cv) for the family of Gaussians in the measurement distortion matrix is determined from an analysis of the sums of squares of fits of the computed DNA histogram to the real data over an appropriate range of trial cv. Since this method assigns a Gaussian to each and every data channel, it permits the resolution of closely spaced multiple aneuploid GI peaks in clinical samples. The effects of high frequency noise that may be present in the data can be attenuated by multiplying the real data histogram by a Gaussian matrix with cv close to but smaller than that of the measurement distortion matrix. the assumption of a single GI peak to constrain the analysis and facilitate the S fraction calculation.In this paper we describe a new iterative method for enhancing the resolution of measured DNA histograms. Our method models the inverse of the Gaussian measurement distortion process without imposing constraints on the curvefitting procedure that would interfere with the detection of aneuploid GI peaks. Since our method assigns a Gaussian function to each data channel, it is capable of resolving multiple aneuploid peaks that are closely spaced. A method for dealing with high frequency noise and its effects on data analysis is also described. Formulation of the ProblemThe DNA histogram obtained by flow cytometry can be considered as the product of an ideal DNA histogram column vector and a measurement distortion matrix. The elements of the column vector are equivalent to the fractions of cells in successive channels of the ideal DNA histogram. The meas-376

A multibeam optical detection system has been developed with a high optical efficiency, achieved through a reduction in the number of optical interfaces employed in the system. This reduction is made possible by a combination of employing simple lenses, gluing the objective lens directly upon the face of the flow cuvette and the extraction of only one fluorescence signal from each laser beam. A This report describes the design and construction of a flow chamber and associated optical fluorescence detection components that permit the simultaneous use of up to three laser beams for flow cytometric investigations. By employing more than one laser beam, different fluorochromes may be excited sequentially with a spatial separation between excitation locations. This permits separate apertures to be used for each fluorochrome, thus reducing the crosstalk between channels and providing for an increased number of analytical options for the investigator (1,2,5,6,8,9). The new design makes efficient light collection possible by gluing the objective lens directly upon the flow cuvette (4) and by extracting only one fluorescent signal from each laser beam employed. Although the Los Alamos system (7) that we had been using included electronic cell volume capability, we had experienced difficulties with the extensive amount of electrical isolation and shielding required for satisfactory operation. In the new design the shielding and isolation requirements were minimized by the incorporation of fluidic electrical isolation resistance elements located in a small shielded enclosure mounted on top of the flow chamber.

An adaptive count-dependent algorithm for smoothing statistically limited histograms has been developed. It considers both the spatial frequency limitations of the measurement system (described by the measurement system point spread function) and the reliability of the measured data (indicated by the effective number of counts influencing each channel of the histogram.Windows for smoothing flow cytometric histograms are derived from an assumed Gaussain-shaped point spread function (PSF) with a constant coefficient of variation. The windows are developed by scaling the variances of the Gaussian functions inversely with the statistical reliability of the data contained in each channel of the measured histogram.The reliability of this data is determined by taking the square root of the number of counts influencing the value tabulated for each channel.Using the algorithm, a smoothed version of the measured histogram may be developed from a linear sum of the products of the individual scaled Gaussian functions and the original measured histogram.Data are presented demonstrating the advantages of count-dependent smoothing over non-count-dependent smoothing using synthesized DNA histograms as a function of sample size.