Kendall Preston

Digital holographic logic

Abstract Theoretically optical computers should be capable of performing 10 14 multiplications per sec. Some, in fact, have attained actual operating speeds of 10 11 multiplications per sec in applications involving data analysis in military radar systems. These computers suffer from the disadvantage of being fixed-program systems, and practical limitations in optical design and optical physics limit precision and accuracy. To take advantage of the high computing rates available in the optical computer and provide accuracy and flexibility optical computers are needed which are (1) digital in nature and (2) functionally programmable. This paper proposes a digital optical computer technology using holographic programming. A simple laboratory demonstration is given where the Identity Function and the Exclusive-OR are generated using this technology. An extension into the practical realm of pattern recognition using acoustically programmed holograms is also mentioned.

A Coherent Optical Computer System Using the Membrane Light Modulator

A 32-input coherent optical computer system is described which is driven by a membrane Light modulator (MLM). The MLM has an active area of 4 by 4 mm containing a 32 by 74 array of 2368 38-pm diameter light modulating elements. The optical computer calculates the one-dimensional Fourier transform over 32 points of a phase function in 1 μs. Both the theoretical and actual operating characteristics of the computer are presented.

AUTOMATIC AUTORADIOGRAPHIC GRAIN COUNTING USING THE CELLSCANTM SYSTEM

With the advent of the CELLSCAN scanning and pattern analyzing system for blood cells, it was evident that an instrument had been devised with imaging capabilities significantly exceeding those of other such systems. For example, FIGURE 1 shows the modulation transfer function (MTF) of the CELLSCAN imaging system. This system consists of a monochromatically illuminated Lietz Ortholux microscope using a 1.3 numerical aperture oil immersed objective and a three times projecting eyepiece connected to a modified Dage slow scan television system. As can be seen from FIGURE 1 , the 10% cutoff of the MTF is at approximately 3,000 cycles to the millimeter. This extremely high resolution made i t apparent that, although the CELLSCAN System was originally designed for transmitting images of white blood cells to a computer for image analysis, it might also be possible to utilize the high resolution achieved for the purpose of imaging grains in autoradiographs. As is well known, the grain size in autoradiographs ranges from approximately two tenths of a micron to almost one micron. FIGURE 2 shows an early result of applying the CELLSCAN system to autoradiographic images. This figure shows a heavily labeled white blood cell imaged by the CELLSCAN television microscope at two different wavelengths. An illuminating wavelength of 4,800 Angstroms was utilized to provide high contrast on the grains and a wavelength of 5,600 Angstroms was utilized to produce high contrast on the entire nucleus. The two images in the lower half of FIGURE 2 are the binary or “quantized” images that are delivered to the computer. As described by Dr. M. Ingram in this Anna1 the CELLSCAN computer counts and sizes black regions in the quantized image. In the case of autoradiographic grain counting it counts and sizes both individual grains and grain clusters. With a modulation transfer function cutoff at 3,000 cycles to the millimeter the CELLSCAN image of a 20 X 20 micron field contains 14,400 data points. The effective line-to-line spacing on the vidicon is 0.1 micron so that actually 40,000 sample points are taken by the scanning mechanism. In order to reduce this enormous number of sample points to a number more reasonable for subsequent computer image data processing, the number of data points is reduced before storage in computer memory. The method utilized for data reduction is in the link between the CELLSCAN scanning system and the CELLSCAN computer. Its function can be defined as a local operator wherein nine points in each 0.3 X 0.3 micron region are examined to extract a single binary value. The result of this examination is a data reduction to an image of only 63 X 63 data points in the computer memory. These data points still retain the basic meaning in the original image. For CEL,LSCAN operations with white blood cells any sample point in an 0.3 X 0.3 micron region being a hinary 0 causes the stored data point to be a binary 0. For autoradiographic grain counting this local operation is unsatisfactory because all grains less than 0.3 microns in diameter are rejected. It

ADVANTAGES OF TOPOLOGY AS A BASIS FOR AUTOMATIC ANALYSIS OF BLOOD CELL IMAGES

Blood cell classes presently of interest to clinicians and to investigators engaged in hematological research were originally defined on the basis of morphology, i.e., shape and size as well as color and staining intensity of cells, nuclei, nuclear lobes and appendages, cytoplasmic granules, and so forth. All of these morphological characteristics have physiological implications. Indeed, the relative numbers of the various classes of leukocytes is onlv one of many types of information sought by the hematologist when he examines a blood smear. It is becoming increasingly clear that there is an acute need for a practical instrument capable of analyzing blood cell images on the basis of widely accepted morphological characteristics. It is, of course, possible to classify blood cells according to characteristics other than morphology, for example, volume, electrical properties, density, presence or concentration of various chemical constituents, and so forth. The feasibility as well as the medical usefulness and significance of such classifications, however, would have to be convincingly demonstrated, and this would require a separate major research effort. In contrast, the information inherent in cellular morphology is widely recognized; the need is for methods to exploit it more fully. This report is intended to review the types of information that are useful for classification of blood cells and that have diagnostic or prognostic significance.

On Determining Optimum Simple Golay Marking Transformations for Binary Image Processing

A computer program has been written which is capable of evaluating all possible Golay marking transformations related to a particular pattern recognition task using an exhaustive search technique. Such evaluations are reported for separating two image data sets: one taken from biomedical microscopy; the other, aerial reconnaissance.

IMPORTANCE OF AUTOMATIC PATTERN RECOGNITION TECHNIQUES IN THE EARLY DETECTION OF ALTERED BLOOD-CELL PRODUCTION

Although it appears that many hematopoietic disorders may be either explosive or relatively insidious in onset, and persistent, progressive, or abortive in development, there is, in fact, little definitive information about preclinical stages of the disorders. Consequently, methods for identifying and studying the earliest stages of aberration have important implications with respect to early diagnosis, as well as pathogenesis. With reference to abnormalities in blood-cell production, quantitative evaluation of the incidence of rarely occurring unusual cell types and of slight morphological changes in large number of commonly occurring blood cells represent sources of information that are readily accessible but presently unexploited because these variables are not susceptible to quantitative evaluation by classical microscopy. A recently developed pattern recognition system for rapid automatic analysis of cellular morphology may make it feasible to use this large body of information about blood-cell production. The present discussion is concerned with the diagnostic implications of an automatic system which may be capable of inspecting the blood-cell “production line” by identifying and counting various types of blood cells as well as aberrant forms that serve as clues to early, subtle derangement of hematopoiesis. Studies on the incidence of rarely occurring cell types are, in a sense, analogous to classical Papanicolaou studies of cells exfoliated from tissues other than blood. Although blood is one of the easiest tissues to sample, and is sampled almost routinely in clinical studies, quantitative studies of the rate of occurrence of unusual cells of hemic origin remain largely undeveloped. The reason for this is that blood is such a large, diffuse tissue and contains so many cells per unit volume, that unless enormous numbers of abnormal cells are produced, they are so quickly diluted, so widely dispersed, that the probability of finding rare abnormal forms in the minuscule samples usually examined becomes very small, indeed. If production line errors in the form of abnormal cells occur extremely rarely, tens or hundreds of thousands of their normal counterparts may have to be examined before finding them. The search then becomes analogous to the proverbial search for the needle in the haystack. If the significance of the rare cells requires not merely knowing whether they are present or absent, but knowing also their true rates of occurrence, the difficulty is compounded. Diagnostic implications of minor degrees of morphological alteration in red blood cells are also largely unexploited. In various types of disease there may be irregularity in size, shape, or hemoglobin content of a greater or lesser proportion of the cells. When the abnormality is severe and a large percentage of the cells demonstrate the change, there is no difficulty in recognizing it, even