Coherent Optical Data Processing Systems Research Articles

Screening of cervical cytological samples using coherent optical processing Part 1

Coherent optical signal processing methods for screening Pap smears were evaluated and are presented in a three-part sequence. In Part 1, 2-D Fourier spectra of normal and abnormal cells generated from many high resolution cell photographs are presented. Each cell spectrum was measured with a coherent optical data processing system containing a special geometry detector and automated data collection capability. Several parameters, determined from weighted measures of the cell transform intensity variations, were tested as feature discriminators to separate normal from abnormal cells. An analysis of the experimental data demonstrates that several transform features are good discriminators of normal/abnormal cells (using standard Baysian decision algorithms with quadratic decision rules). In Part 2, mathematical model studies to guide and validate the experimental work show certain transform parameters to be functionally related to cell and nucleus diameter and optical density. Other parameters appear to be related to cell characteristics such as clumping of nuclear deoxyribonucleic acid (DNA). The model studies also show that the photographic variables play a key role in cell image preprocessing prior to Fourier analysis. Part 3 discusses an optical transducer that was used as a film replacement to modulate a laser beam spatially with a cell image. Several of the feature discriminators used in Part 1 with photographic film images served also to separate normal and malignant cell types when the cell Fourier spectrum was obtained from a transducer image. Based on an understanding of the procedure from the model studies and the demonstrated ability to separate normal and malignant cells using certain transform features, a coherent optical processing system to screen Pap smears from cell or slide photographs is feasible and appears practical in terms of the number of cells to be processed. A high-speed optical transducer would be required for processing large numbers of cells without photography in a reasonable time interval.

General theory of linear, coherent optical data-processing systems

This paper concerns a general theory of linear optical data-processing systems. Well-known basic ideas are critically reviewed, fundamental constraints upon the system behavior are investigated, and a new operational notation is proposed. Every system is considered as a bidirectional processor and every two-dimensional signal is equivalently described in the space and frequency domain. Without referring to field-theoretical notation, the constraints of “reciprocity” and “losslessness” are introduced. In rough terms, reciprocity states equality of transmissions A → B and B → A, where A and B are points in the two reference planes, losslessness states equality of signal energies in the two planes. “Symmetry” has to be independently defined in the space and the frequency domain, and “shift invariance” and “spreadlessness” appear to be dual constraints. The important constraint of “time reversibility” is obtained when a system simultaneously satifies reciprocity and losslessness. In the second part, the general theory is applied to lenses, sections of free space, Fourier transformers, and magnifiers. With the aid of a suitable shorthand notation, the main results of Fourier optics can be easily derived.

Application of binary filters in automated counting of reticulocytes.

Computer generated binary matched filters have been used in an attempt to recognize and count immature red blood cells (reticulocytes) on a standard blood smear slide in a coherent optical data processing system. The Fourier plane binary filters were synthesized on a digital computer using an appropriate polar geometry mathematical model. Various filters were constructed and tested for synthetic blood cells, and preliminary test results are presented for random samples of actual blood slides. These show a trend of manual reticulocyte count versus the matched filter count, but no statistical significance.

Coherent optical recognition and counting of reticulated red blood cells.

A coherent optical data-processing system for identifying and counting blood cells has been developed. Optical Weiner-Kolmogorov (minimum mean-square error) linear filtering is employed to estimate the density per unit area of a given cell type present in a sample blood smear. The Weiner filter is designed to recognize the characteristics of an ensemble of cells of a type rather than specific individual cells. In this way the system is made insensitive to minor variations in shape, size, or orientation of the cells. A Poisson shot-noise model for blood cells as a spatial random process is employed to calculate the far-field scattering from the cells. Results of experiments in which the percentage of reticulated (immature) red cells present in a smear has been determined are presented. The identification and counting accuracies compare favorably with those obtained by laboratory technicians with a microscope. The potential time saving of instantaneous estimation for the entire smear is significant.

Synthesis of a Spatial Filter for the Combined Operations of Subtraction and Correlation

As is known, a coherent optical data processing system can be used to perform convolution and correlation with ease. Recently it has been shown that the system with diffraction gratings in the spatial filtering plane can also add or subtract complex patterns displayed symmetrically in the input plane. A method is described here for synthesizing a spatial filter so that the coherent system can perform both correlation and subtraction simultaneously in real time. As a particular application of this technique, we present results of using such filters in optical feature extraction.

Resolution Testing of Coherent Optical Systems by Means of Linear Gratings

Testing of coherent optical data processing systems has shown a need for a method which measures resolution at all points in the image format. The use of multifrequency linear gratings as test objects satisfies this need. Quantitative data may be obtained if care is taken to calibrate and linearize the photographic process. However, the method's principal application seems to be for qualitative testing. A comparison is made between calculated wavefront aberrations and experimental results obtained using linear gratings.

Linear Vector Operations in Coherent Optical Data Processing Systems

The generation of a two-dimensional linear vector space in a coherent optical data processing system and its specific application to the gradient operator is discussed using theoretical and experimental results. The vector space is created by superposing the outputs of two Fourier optical systems having light of mutually orthogonal polarizations. As a result, the total amplitude of the signal in the output plane is a vector sum of the signals from the systems. If each one of the systems performs one of the partial derivative operations of the transverse gradient, and if the inputs to both systems are identical, then the output is the vector sum of the partial derivatives or the transverse gradient operation. The experimental program is heavily oriented toward the realization of the optimum approximation to the jomega(x)î and jomega(y)j; filters, rather than using various binary-type filters. Problems with film and lens noise are also discussed.

Signature recognition for nuclear detonations.

Experiments have been performed on the recognition of nuclear detections by using coherent optical data-processing systems. The data, E-field signals recorded with a 10 khz bandwidth, were available from only ten events, all airdrops. Since the signals were of an unknown form, they were classified into sub-groups by using a cross-correlation technique. Spatial carrier-frequency filters, constructed for these signals, were then tested to determine their ability to recognize all ten signals while rejecting noise from atmospheric sources. The experiments, although of a preliminary nature, show that all the test signals can be recognized with a two-channel filter and that the nearest noise peak (from six samples) is 6–8 db less than the minimum correlation peak from any of the ten signals.