Five leukocyte analyzers (three from the United States and two from Japan) entered commercial productionduring the decade 1974-1984; and three (all from the United States) entered pilot production, but wereabandoned before complete commercialization. These instruments are automated microscopes, most ofwhich have robotic capabilities in handling glass microscope slides. All incorporate machine vision comput-ers for the purpose of performing multispectral image processing and pattern recognition. Since the objectswhich these machines must recognize (human white blood cells, red cells, and platelets) are small (from 2 to 20jum in diameter) and have significant microscopic detail, the robotic manipulation of the microscope slidemust be extremely precise, and the accompanying electrooptical imaging system must furnish submicronresolution. Today, these leukocyte analyzers worldwide analyze the cellular blood constituents of tens ofthousands of patients per day. They gather clorimetric and morphometric data on about one billionmultispectral images per year and compute and report their analysis of this data. Unfortunately, the fifty-million dollar investment made collectively by the business community has led to such a disappointing returnthat, as of 1986, all United States manufacturers have abandoned production. At the present time, only twomachines are in production, both in Japan. This paper describes the historical evolution of these leukocyteanalyzers and furnishes technical detail on the electrooptical characteristics of two major United Statesmachines.
1. Introduction
The history of human leukocyte (white blood cell)analyzers is basically the history of the application ofmachine vision and robotics to optical microscopy. Itsorigins can be traced to a committee meeting of theNational Coal Board (Great Britain) chaired by J.Bronowski, convened in 1951, to investigate "the pos-sibility of making a machine to replace the humanobserver."' Although the quantitation of aerial coaldust was the driving force behind this effort in automa-tion, the real leadership came from bioengineers in theDepartment of Clinical Pathology, Radcliffe Infirma-ry, Oxford, the Department of Anatomy, UniversityCollege London, and the Medical Research Council.Within a year of the meeting, the flying spot micro-scope of Young and Roberts2 had been built. Thisinstrument was employed in counting and sizing hu-man red cells at 1-Am resolution and, using simplepulse-length association circuits, generating a size his-
The author is with University of Pittsburgh, Pittsburgh, Pennsyl-vania 15213.
togram (0-30 ,m in 5-Am intervals) of all red cells in a500 X 500-,um field in 4 s. The difference betweenmeasurements made by machine and observer for se-lected fields was reported as 42%. Commercializationof this instrument took place on a small scale by suchcompanies as Mullard and C. F. Cassela, Ltd.
II. Blood Cell Analysis
There is one white blood cell for approximately ev-ery one thousand red blood cells in human blood.There are five major categories of white blood cells(Fig. 1). Classification of white cells into these majorcategories provides a valuable indicator of many dis-eases. This enumeration is called the white blood celldifferential or, simply, the WBCD. The standardWBCD is conducted by collecting blood (either by skinpuncture or venipuncture), spreading a few drops ofthe specimen on a glass microscope slide, staining itwith biochemical dyes (ordinarily methylene blue andeosin) and examining it for a sufficient length of time(5 or 10 min) to determine the types of 100 white cell(while simultaneously screening 100,000 red cells).At the same time, the number of platelets (cells in-volved in blood clotting) are screened for an estimateof what is called platelet sufficiency. The technologistwho performs this task observes the color, morphology,and uniformity of red cells and platelets while differ-entially categorizing the 100 white cells. Details ofthis operation have been provided by the College of
Fig. 1. Artist's rendition of the major human white blood cellsshowing their salient morphometric characteristics' (monocyte, eo-sinophil, lymphocyte, basophil, and nonfilamented and filamentedneutrophils) interspersed with red cells and platelets. In actualsmears of human blood, approximately one white cell is found for
each thousand red blood cells.
American Pathologists Workload Recording Commit-tee.3
The blood count and routine urine analysis are thetwo examinations most commonly specified for pa-tients around the world. The WBCD is one compo-nent of blood analysis which provides nonspecific, i.e.,confirmatory, information on many diseases. For ex-ample, neutrophilia may confirm the presence of aninflammatory abdominal disease such as appendicitis;eosinophilia, parasitic or allergic disorders; basophilia,myeloproliferative diseases; lymphocytosis, viral in-fections. The presence of blasts (immature whiteblood cells) may indicate leukemia. The presence ofnucleated red cells (red cells which ordinarily appearin the blood stream without nuclei) may indicate mye-lophthisic anemia. Since the WBCD is nonspecific,shows diurnal variations, and is subject to samplingerror, its usefulness has sometimes been questioned.4-6Despite these criticisms, however, the WBCD is car-ried out in -50% of all blood analyses requested. Thecurrent estimate is that 150 million WBCDs are nowdone each year in the United States alone. Thus, thisparticular segment of the health services industry isestimated to produce a gross revenue of about twobillion dollars per year.
iii. First Reduction to Practice
With the demonstrated success of high-speed redblood cell counting by TV microscopy in Great Britain,the United States Atomic Energy Commission (nowpart of the Department of Energy) decided to contractwith the Perkin-Elmer Corporation (Norwalk, CT) fora feasibility study on automating white blood cell dif-ferentiation for the purpose of locating binucleated
Fig. 2. The above is the only known photograph of the CELLSCAN/
GLOPR system" which performed the first reduction to practice of anautomated WBCD. The system included (1) microscope illumina-tor power supply, (2) automatic focus mechanism, (3) automaticmicroscope, (4) microscope slide traverse mechanism, (5) white cellfinding and framing computer, (6) teletype, (7) image recognition
computer, and (8) commercial minicomputer.
lymphocytes. (It had been shown manually that theincidence of the binucleated lymphocyte increased sig-nificantly in those individuals who were exposed tolow-level ionizing radiation.) This led to the construc-tion of the CELLSCAN system under Preston.7 Undersponsorship of the United States National Institutes ofHealth and the Internal Research and DevelopmentProgram, sponsored by the Department of Defense,Preston and his colleagues were encouraged to broadentheir study to include differentiation of lymphocytes,monocytes, and neutrophils. This was accomplishedby CELLSCAN/GLOPR (Fig. 2), as reported by Ingramand Preston.8 Although CELLSCAN/GLOPR was slowand cumbersome (an overnight run from 5:00 in theevening to 7:00 the next morning was required for each500-cell count), it was described by Neurath9 as a"landmark which points the way."
Both the original CELLSCAN and CELLSCAN/GLOPRused the Leitz Ortholux microscope with a 10OX, 1.30N.A. objective operating under oil immersion.Whereas CELLSCAN used a Dage Data-Vision, slow-scan TV camera operating at 60 horizontal lines/s andsynchronizing the associated binary image processingcomputer, CELLSCAN/GLOPR employed an oscillatingmirror scanner using an RCA 8645 ten-stage photo-multiplier with a horizontal scan rate of 40 Hz whichtransmitted integer image data. CELLSCAN/GLOPRemployed a finding and framing special-purpose com-puter for locating white cells and, subsequently, a Var-ian 620i minicomputer for image storage (16K bytes ofmain emory or 7-track magnetic tape) and sub-sequent analysis by the Golay logic processor(GLOPR). A detailed discussion of CELLSCAN is giv-en in Izzo and Coles10 and of CELLSCAN/GLOPR byPreston."1 Within a year after the first demonstra-tions by CELLSCAN/GLOPR, the instrument had beenreprogrammed to recognize all five major types of hu-man white blood cells.12
Fig. 3. Commercial robot microscopes for performing human blood cell image analysis by high-resolution multispectral morphologicalpattern recognition. These are (A) The SmithKline Beckman HEMATRAK, (B) the Coulter Biomedical Research Corp. diff3 (designed by
the Perkin-Elmer Corp.), (C) the Hitachi 806, and (D) the Omron Tateisi Electronics Company Microx Cell Analyzer.
IV. Commercialization
Following the publication of Ingram and Preston,8research at the University of Pennsylvania by Miller'3led to the establishment of the Geometric Data Corpo-ration (now a subsidiary of SmithKline Beckman) andthe manufacture and marketing of the Hematrakproduct line. The first two installations of this instru-ment were in the clinical hematology departments ofAndrews Air Force Base and the Greater SoutheastCommunity Hospital (Washington, DC) in December,1974. At the same time, research at the University ofChicago by Bacus14 led to the development of anotherinstrument-the LARC, manufactured by CorningGlass Works. By the mid-1970s, about 50 LARCs andHematraks were in clinical use.
By 1976, the Instrument Group of the Perkin-ElmerCorp. had constructed an instrument called the diff3,based on CELLSCAN/GLOPR. Faced with problems inobtaining FDA approval of modifications of this in-strument under the Medical Device Amendment of197615 and the difficult problem of developing a salesand maintenance force in an area foreign to their mainline of business, Perkin-Elmer transferred the entireproduct line to Coulter Electronics in 1977. Coulter
established a manufacturing site in Massachusetts,namely, the Coulter Biomedical Research Corp.
Prior to this, Coulter had established the CognosCorp. in 1979 and had brought the Coulter differentialinto pilot production, whose design was based on re-search by Young16 at the Massachusetts Institute ofTechnology. This effort had been unsuccessful. An-other failed effort was that of Abbott Laboratorieswhich built the ADC-500 based on the work of Green.17
The performance of both the Hematrak and thediff3 analyzers was improved by the production of newmodels. Coulter designed the diff3/5018 and later thediff4,19 although the latter was recalled. In Japan,both Omron and Hitachi began manufacture of theirdifferential counters. Some of these instruments areshown in Fig. 3.
Extensive evaluations of the major differentialcounters were made by various clinical groups.2 0 -2 3
The results of these studies indicated that these differ-ential counters were equivalent in their accuracy inidentifying normal white blood cells from their imagesto a well-trained technologist. (Some of these ma-chines stored coordinates of abnormals which couldsubsequently be reimaged for visual scrutiny by thehuman expert.) At 4000 white blood cells analyzed
per hour, a technologist using the automated differen-tial counter could perform the work of four technolo-gists using manual methods.
Smith Kline Beckman calculated that a technologistcould readily inspect 40 human blood films/h using theautomated differential counter, but only 10/h manual-ly. With expenses (at that time) of $12.25/h, this givesa labor cost savings of $0.92/specimen. For an instal-lation averaging 120 films/day (365 days/year), thisyields an annual savings of $40,296 with a 2.7-yearpayback on a $90,000 machine (including the $7,200annual service contract). As the cost of technologistlabor has increased and, more importantly, with thedifficulty of training technologists in the tedious workof performing the WBCD, machine vision systems forhematology were employed by a large number of themajor hospitals in the United States by the end of the1970s. At that time, there were -1000 robot micro-scopes in use, so that automation of the leukocytedifferential represented the major application of ma-chine vision in medicine.
V. Technical Details
Table I provides a tabular comparison of the Hema-trak and the diff3. Major differences can be observed.Hematrak employed a flying spot scanner which di-rectly scanned the microscope slide and generatedmeasurements as objects of interest were scanned. Onthe other hand, diff3 used two scanners (both of whichwere of the imaging type). A solid-state line scannerwas used to search the field for objects of interest,which were then centered and reimaged by means of aplumbicon TV tube. Focus was not continuous andoccurred in discrete steps after each object of interestwas located. Feature extraction in Hematrak was byspatial-colorimetric probability density functions,whereas in diff3, more traditional measures of size,shape, and color were executed. The exact measuresgenerated by Hematrak have never been published.Those generated by diff3 are given in Table II.
Despite the dissimilarities mentioned above, thegeneral performance of these two instruments is suffi-ciently similar so that a discussion of the electroopticalcharacteristics of one should provide the reader withsufficient background to understand the overall opera-tion of the robot WBCD microscope. The descriptionwhich follows is taken to a large extent from the excel-lent paper by Graham and Norgren.18
The optical schematic for the diff3 is shown in Fig. 4.A 60-W tungsten lamp is used with a dry 0.9-N.A.condenser using Kohler illumination to provide lightto the microscope slide. The objective is a 40X, 1.0-N.A. oil-immersion planapochromat which relays theimage of the slide along three optical paths. The firstgoes to the visual oculars which allow the operator toexamine the field of view in white light through 12.5Xwide-field oculars mounted above a variable-magnifi-cation relay (Optovar), which gives overall magnifica-tion of 50OX, 625X, 800X, or 1000X. The second opti-cal path goes to the solid-state line scanner via a 63-
mm photoprojection ocular whose magnification issuch that the line scanner searches a 200-,um swath onthe slide. A 50 X 50-,Mm object field is relayed to theplumbicon TV camera which is placed at the focalplane of another 63-mm photoprojection lens. At theexit pupil of this lens is mounted a galvanometer-driven mirror whose purpose is to center objects locat-ed by the line finder. Also at this position is a sema-phor carrying filters centered at 515 and 580 nm. Im-age data taken through these filters are used for focuscontrol, image centering, and object classification.
The field of view is digitized at 6 bits/picture ele-ment in a 128 X 128 format having 0.4-,um pixel spac-ing. The TV camera is essentially a special-purposeperipheral to the control minicomputer (Data GeneralNova model 4) which, by software control, stores eithera 64 X 64 field at 0.8-,Mm spacing or one-quarter of the128 X 128 field at 0.4-,Mm spacing. Data acquisition fora single image frame takes place in 10.6 ms.
The entire optical assembly of the diff3 is mountedon a shock-mounted optical bench designed for therigorous environment of the clinical laboratory. Allimaging optics are carried by a cast aluminum bridgeto which the line scanner and plumbicon camera aremounted. The xyz mechanism for handling the mi-croscope slide is carried on a diaphragm flexure whichis free to be driven in the z direction by a coarsefocusing mechanism.
Connecting a linear bearing and the diaphram flex-ure is a cup which surrounds the condenser optics.This allows the condenser to remain stationary whilethe diaphram is flexed using a cam which contacts thebearing shaft. This is driven by a stepping motorwhich yields coarse-focus steps of 0.8 um (with a rangeof 500,Mm). A piezoelectric element drives the cantile-vered microscope slide holder mounted to the mechan-ical stage in steps of 0.12 gm (over a 7 .5 -,um range).The x-y stage has ball bearing ways and is driven ineach axis by a precision lead screw and stepping motor.Positional resolution is 3.2 m (over a range of -1.0 X2.0 cm). Microscope slides are autoloaded from acassette using a robot arm which removes the slidefrom the cassette, inserts it into the slide holder, and(when analysis is complete) delivers it back to thecassette. The cassette capacity on the diff3 is fourteenbeveled hematology slides.
Feature extraction in the diff3 is carried out usingthe Golay parallel pattern transform, 2 4 which is de-scribed in some detail by both Graham and Norgren18
and Preston and Duff.25 The particular Golay logicprocessor (GLOPR) employed in the diff3 conducts a 3X 3 binary convolution in 25 ns (40 million picturepoint operations/s), which permits a single Golaytransform over the 64 X 64 field in 100 us. This isaccomplished by using eight processing elements inparallel, each of which performs its calculations bytable lookup. The GLOPR is microprogrammed fromthe Data General Nova host using a 48-bit microcodeword which indicates (1) source and destination regis-ters, (2) the Boolean function to be performed, (3) theGolay hexagonal patterns on which to trigger, (4) the
number of iterations for which the given transform isto be performed, and (5) the subfield specification (ifrequired). Total time necessary for executing all mea-sures given in Table II is 200 ms. The entire cycle offinding an object of interest, centering it in the field,focusing, transferring image data to the host, generat-ing all measures, and carrying out object identificationthat takes place in somewhat <1 s. Thus, the diff3 is
the fastestworld.
multicolor machine vision system in the
V. Present and Future Prospects
Despite tremendous technical accomplishments bythe corporations which have brought the robot WBCDmicroscopes to the market, these instruments (as withmany other instruments applying machine vision to
Green cytoplasm transmissionRed cytoplasm transmission
TELEVISION VISUALING CAMERA PATH
OPTOVAR SCNE
OCULAR OCULAR
Pare + ~CELL IDE
OBJECTIVE -43
SLIDE I T UN STE
CONDENSER ~S U C
EN
Fig. 4. Optical schematic of the Coulter diff3 showing how opticalimages are formed simultaneously for visual analysis, high-speedwhite blood cell finding, and multispectral high-resolution televisionimaging. Once each white blood cell is found, centered, and imaged,both red and green image data are digitized into two 64 X 64 X 6digital memories. These data are histogrammed at the full videorate, thresholded, and the resultant binary images stored within theGolay image processor. This processor requires 200 ms to extractapproximately fifty clorimetric and morphometric features andidentify the white blood cell. The resulting WBCD is printed on the
patient's record.
robotics) have not proved to be economically success-ful. However, this new technology has brought accu-racy, uniformity, reproducibility, and new levels ofquality control to performance of the WBCD.
This is due not only to the robot microscopes de-scribed above, but also to automation of the handlingof human blood and preparation of uniform, monolay-
er films on microscope slides. It has transformed anindustry previously characterized by the inaccuraciesresulting from the tedium of visual inspection(Sencer26 of the Centers for Disease Control claimedthat 40% of all manual WBCDs were outside of threesigma limits) into an entirely new world of automationwith all its concomitant benefits. On the other hand,the -50 million dollars of investment (United Statesonly) has resulted in only -200 million dollars of grossrevenue over a period of -10 yr. In fact, as of 1986, allUnited States manufacturers had ceased production.Coulter Biomedical Research Corporation closed itsmanufacturing facility in April 1984. In October of1986, the Geometric Data Division of SmithKlineBeckman announced that they would stop productionof the Hematrak line. Only Omron and Hitachi con-tinue to make robot WBCD microscopes. Beard27
quotes Wallace Coulter as saying, "We lost severalmillion a year making, selling, and maintaining thesemicroscopes. The problems turned out to be bigger;and the market, smaller than all of us in businessthought. We believed that, since the microscope issuch a universal tool, an automated microscope wouldfind a place. It hasn't yet."
Another blow to automation of the WBCD were theDiagnostic Related Reimbursement federal regula-tions introduced in 1984, which put economic pres-sures on hospitals to justify all clinical laboratory test-ing. Consequently, many United States hospitalsbegan doing WBCDs on admissions only. This result-ed in an -30% reduction in WBCDs. Another majoreffect was to make expansion of robot microscopy intomid-size hospitals economically unjustifiable withouta major reduction in selling price of these instruments(from -$150,000 each to, say, $50,000).
Due to FDA rulings,'5 owners of diff3 and Hema-traktype differential counting microscopes who wishto retain their instruments will be guaranteed serviceand maintenance through August of 1987 and Octoberof 1991, respectively. Thus, by the 1990s, there will bea loss of the 1000 instruments manufactured by thesetwo organizations. Whether Omron or Hitachi willcontinue to manufacture and, eventually, replace in-struments lost is unknown at the present time. Thus,at this writing, the future of these leukocyte analyzersis uncertain at best.
These discouraging circumstances in the industrialfield do not imply that interest has been lost in theapplication of machine vision to the WBCD. Re-search still continues. Ph.D. theses on this subjectwere completed recently at the University of Groning-en (see paper by Bins et al.2 8) and at the University ofAmsterdam by Landerweed,29 while Zajicek et al.3 0 inIsrael continues to pursue studies on nucleated redblood cells. In Germany, Aus et al. 31 is performingresearch on high-resolution images of abnormal bloodcells. Whether this will lead to the introduction of aEuropean instrument is unknown, although there havebeen some indications in this direction as witnessed bythe SAMBA project32 in France. In the United States,Tufts University is studying diagnostic indicators in
peripheral blood smears. This has been reported byZahniser et al.3 3 3 4
This paper would be delinquent if it did not includesome discussion of competing technology. The redblood cell counters of Mullard and Cassela were re-placed in the 1960s by the fluid-flow Coulter counter.35
We find also that competing fluid-flow technology hascontributed to the downfall of the WBCD robot micro-scope. As described in two special issues of thejournal-Blood Cells,3 6 fluid-flow WBCD analyzers by Coulter,Ortho, and Technicon, although not capable of differ-entiating as many white cell classes as the diff3 andHematrak, are rapidly approaching this capability dueto the development of new immunologically based bio-chemical reagents which tag the different classes ofhuman white blood cells. These machines have theadvantage of processing body fluids in a fluid mediumand of passing some 5000 cells/s through an interroga-tion orifice. However, the disadvantage is that abnor-mal cells, once detected, are not readily seen by thehuman eye. In a screening environment, where mostbloods are normal, the fluid-flow machine may be theideal. They accurately recognize the rare abnormalblood, which is then visually screened using the tradi-tional microscope slide blood smear. As with the ro-bot WBCD microscopes, these machines alleviate thetedium of visually screening normals and use the skill-ful eye-brain system of the technologist for the moredifficult task of categorizing abnormal cells.
In summary, the past quarter century has seen re-markable technological developments in the robotiza-tion of the WBCD microscope, with clear demonstra-tions that these machines can screen red cells at100,000/min, locate the accompanying 100 white cellsfor detailed scrutiny at high magnification, and pro-duce, at least for normals, an analysis superior to thatgenerated by human vision. This is a signal accom-plishment which, unfortunately, could not be support-ed economically, at least by United States businesses.Although the future holds some promise for robotiza-tion of microscopy, it will be difficult to justify furtherindustrial efforts in this field unless the cost of theseinstruments can be significantly reduced to becomecost-competitive with fluid-flow systems and theirpattern recognition capabilities can be augmented toinclude the differentiation of, at least, the most com-mon forms of abnormal cells.
The author wishes to thank N. Pressman, editor ofthis special issue, for encouraging the production ofthis paper. In addition, the author would like to giveparticular thanks to the several reviewers, namely, M.D. Graham, Coulter Biomedical Research Corp., P. E.Norgren, Perkin-Elmer Corporation, M. N. Miller,Numar Corporation, and J. A. Koepke, Duke Universi-ty Medical Center, who provided immeasurable assis-tance in suggesting improvements to the manuscript.Finally, the assistance of Monica Williams and CathyMagdaleno, Executive Suite, Tucson, is much appreci-ated in the preparation of the text and tables.
The author also holds a professorship at Carnegie-Mellon University.
References1. W. H. Walton, "Automatic Counting of Microscopic Particles,"
Nature 169, (1952).2. J. Z. Young and F. Roberts, "A Flying Spot Microscope," Nature
167, 231 (1951).3. Manual for Laboratory Workload Recording Method, College
of American Pathologists Workload Recording Committee(College of American Pathologists, Skokie, IL, 1987).
4. H. Uhrbrand, "The Number of Circulating Eosinophils: Nor-mal Figures and Spontaneous Variations," Acta Med. Scan. 160,99 (1958).
5. C. L. Rumke, "The Statistically Expected Variability in Differ-ential Leukocyte Counting," in Differential Leukocyte Count-ing, J. A. Koepke, Ed. (College of American Pathologists, Sko-kie, IL, 1978), pp. 39-45.
6. E. C. Rich, T. W. Crowsin, and D. P. Connelly, "Effectiveness ofDifferential Leukocyte Count in Case Finding in the Ambula-tory Care Setting," J. Am. Med. Assoc. 249, 633 (1983).
7. K. Preston, Jr., "The CELLSCAN System: A Leukocyte PatternAnalyzer, in Proceedings, Western Joint Computer Conference(Institute of Radio Engineers, New York, 1961), pp. 173-177.
8. M. Ingram and K. Preston, Jr., "Automatic Analysis of BloodCells," Sci. Am. 223, 72 (1970).
9. P. W. Neurath, "A Review of Image Processing in Biology andMedicine," Med. Res. Eng. 10, 9-15 (June-July 1971).
10. N. F. Izzo and W. Coles, "Blood Cell Scanner Identifies RareCells," Electronics 35, 52 (April, 1962).
11. K. Preston, Jr., "Use of the CELLSCAN/GLOPR System in theAutomatic Identification of White Blood Cells," BioMed. Eng.7, 266 (1972).
12. K. Preston, Jr., "Applications of Cellular Automata to Biomedi-cal Image Processing," in Computer Techniques in Biomedicineand Medicine (Auerbach, Philadelphia, 1973), pp. 295-331.
13. M. N. Miller and M. S. Levine, U.S. Patent 3,932,687 (August 27,1974).
14. J. W. Bacus, "An Automated Classification of Peripheral BloodLeukocytes by Means of Digital Image Processing," Ph.D. Dis-sertation, University of Illinois, Chicago (1970).
15. Public Law 94-295, Medical Device Amendments of 1976, 94thU.S. Congress S.51028 (1976).
16. I. T. Young, "Automated Leukocyte Recognition," Ph.D. Dis-sertation, Massachusetts Institute of Technology (1969).
17. J. F. Green, "A Practical Application of Computer Pattern Rec-ognition Research: The Abbott ADC-500 Differential Classi-fier," J. Histol. Cytol. 27, 160 (1979).
18. M. D. Graham and P. E. Norgren, "The diff3 Analyzer: AParallel/Serial Golay Image Processor, in Real-Time MedicalImage Processing, M. Onoe, K. Preston, Jr., and A. Rosenfeld,Eds. (Plenum, New York, 1980), pp. 163-181.
19. M. D. Graham, "The diff4: A Second-Generation Slide Analy-zer," in Computing Structures for Image Processing, M. J. B.Duff, Ed. (Academic, London, 1983).
20. J. S. Nosanchuk, P. Dawes, A. Kelly et al., "An AutomatedBlood Smear Analysis System," Am. J. Clin. Pathol. 73, 165(1980).
21. T. C. Kingsley, "The Automated Differential: Pattern Recog-nition Systems, Precision, and the Spun Smear," Blood Cells 6,483 (1980).
22. W. A. Rock, J. B. Miale, and W. D. Johnson, "Detection ofAbnormal Cells in the White Cell Differential: Comparison ofthe HEMATRAK Automated System With Manual Methods,Am. J. Clin. Pathol. 81, 233 (1984).
23. G. S. Cembrowski, M. K. Wilson, S. G. Adelmeyer et al., "Evalu-ation of the Accuracy of the diff3/50 with a Cell by Cell Compari-son," Am. J. Clin. Pathol. 80, 333 (1983).
24. M. J. E. Golay, "Hexagonal Parallel Pattern Transformation,"IEEE Trans. Comput. C-18, 733 (1969).
25. K. Preston, Jr., and M. J. B. Duff, Modern Cellular Automata:Theory and Applications (Plenum, New York, 1984).
26. D. Sencer, Statement to the Senate Subcommittee on Antitrustand Monopoly (Committee on the Judiciary, Washington, DC,1967).
27. J. D. Beard, "Computerized Biopsy System," Comput. News-Physicians 3j C-12 (Mar. 1985).
28. M. Bins, G. H. Landerweed, E. S. Gelsema et al., "Texture ofWhite Blood Cells Expressed by the Counting Densitogram,"Cytometry 1, 321 (1984).
29. G. H. Landerweed, "Pattern Recognition of White Blood Cells,"Ph.D. Dissertation, University of Amsterdam (1981).
30. G. Zajicek and M. Shohat, "On the Classification of NucleatedRed Blood Cells," Comput. Biomed. Res. 16, 553 (1983).
31. H. M. Aus, H. Harms, M. Haucke et al., "Leukemia-RelatedMorphological Features in Blast Cells," Cytometry 7, 365(1986).
32. G. Brugal, C. Quirion, and P. Vassilaka, "Detection of BladderCancer Using SAMBA 200 Cell Image Processor," Anal. Quant.Cytol. Histopath. 8, 187 (1986).
33. D. J. Zahniser, J. F. Brenner, W. D. Selles et al., "DetectingInfection-Related Changes in Peripheral Blood Smears WithImage Analysis Techniques," Anal. Quant. Cytol. Histopath. 5,269 (1983).
34. D. J. Zahniser, J. F. Brenner, and W. D. Selles, "Spectral Band-width in Automated Leukocyte Classification," Cytometry 7,518 (1986).
35. W. Coulter, "High-Speed Automatic Blood Cell Counter andCell Size Analyzer," in Proceedings, National Electronics Con-ference (Institute of Radio Engineers, New York, 1956), pp.1034-1042.
36. B. Bull, Ed., "The White Blood Cell Differential," Blood Cells11, 1 (1985).
Pollution to Prevention, continued from page 3163
Policy Options
Congress faces clear but difficult choices. However,nearly everyone agrees that prescribing waste reduc-tion through regulation is technically infeasible and ad-ministratively impractical. The OTA and EPA reportsto Congress help bring three fundamental policy op-tions into focus:
Policy Option I:Take no new action to directly help industry toreduce waste generation; continue to rely oncurrent industry efforts.
This approach implicitly discounts obstacles towaste reduction that confront nearly all waste gener-ators, such as poor information on the exact sourcesof their wastes and ways to reduce their generation.The valid basis for congressional and public criticismof regulatory programs weakens their positive impactson waste generators. Regulatory programs that are in-effective for their designed purposes are even more in-effective in causing comprehensive waste reduction.Waste reduction does not typically prevail over othertraditional responses to rising environmental costs andliabilities, such as changes in pollution control tech-nologies, acceptance of high and avoidable costs, and,in exceptional cases, plant closings.
Policy Option II:Institute a small Federal effort through existingenvironmental statutes and regulatory programs.
This would limit reduction to certain regulatedwastes, pose administrative problems because of manyother congressionally mandated tasks to EPA, andhave limited credibility because people in existing envi-
ronmental programs are not expert about productionprocesses and have shown little interest in waste re-duction. It might not significantly change what is nowoccurring.
Policy Option III:Through new legislation, establish a separateFederal program within EPA to support wastereduction and to provide national leadership.Fund it and State programs by allocating sev-eral percent of EPA's operating budget.
A nonregulatory approach would address many ob-stacles. It would assist American industry to learn byexperience that reducing the generation of all wastesis technically feasible and in its own economic self-interest to do as soon as possible. A 5-year seed grantsprogram for State efforts could build on existing butlimited State and local programs. Government-fundedin-plant technical assistance and central sources of in-formation, for example, could overcome inertia andsmooth a path from sole dependence on costly end-of-pipe regulations to a dual environmental strategythat includes vollntary, comprehensive waste reduc-tion. Increased corporate profits from waste reductionsavings are likelv to result in sufficiently increased taxrevenues to rapidly offset the cost of a Federalprogram.
Copies of the OTA special report, From Pollution to Pre-vention: A Progress Report on Waste Reduction. " are a-vail-able from the Superintendent of Documents, U.S. Govern-ment Printing Office. Washington, DC 20402-9325 (202)7S3-3238. Te GPO stock number is 052-003-01071-2: theprice is 52. 75. Copies of the report for congressional use areavailable b calling 4-8996.
