The Heidelberg Retinal Flowmeter (HRF, Heidelberg Engineering,

Transcription

1 IOVS, March 1999, Vol. 40, No. 3 Reports Bazan HEP, Varner L. A mitogen-activated protein kinase (MAPkinase) cascade is stimulated by platelet-activating factor (PAF) in corneal epithelium. Curr Eye Res. 1997;l6: Bazan HEP, Tao Y, Bazan NG. Platelet-activating factor induces collagenase expression in corneal epithelial cells. Proc Natl Acad Set USA. 1993;90: Tao Y, Bazan HEP, Bazan NG. Platelet-activating factor induces the expression of metalloproteinases-1 and -9, but not -2 or -3, in the corneal epithelium. Invest Ophthalmol Vis Sci. 1995;36: Tao Y, Bazan HEP, Bazan NG. Platelet-activating factor enhances urokinase-type plasminogen activator (UPA) gene expression in corneal epithelium. Invest Ophthalmol Vis Sci. 1996;37: Izumi T, Shimizu T. Platelet activating factor receptor: gene expression and signal transduction. Biochim Biophys Acta. 1995; 1259: Marcheselli V, Rossowska MJ, Domingo MT, Braquet P, Bazan NG. Distinct platelet-activating factor binding sites in synaptic endings and in intracellular membranes of rat cerebral cortex./biol Chem. 1990;19265:9l40-9l van Delft JL, van Haeringen NJ, Verbeij NLJ, Domingo MT, Chabrier PE, Braquet P. Specific receptor sites for PAF in iris and ciliary' body of the rabbit eye. Curr Eye Res. 1988;7:1063-l Thierry A, Doly M, Braquet P, Cluzel J, Meyniel G. Presence of specific platelet-activating factor binding sites in the rat retina. J Pharmacol. 1989;63: Mori M, Aihara M, Shimizu T. Localization of platelet-activating factor receptor messenger RNA in the nit eye. Invest Ophthalmol Vis Sci. 1997;38: McPherson GA. Analysis of radioligand binding experiment, collection of computer programs for PC. / Pharmacol Methods. 1985:14: Carlson SA, Chatterjee TK, Fisher RA. The third intracellular domain of the platelet-activating factor receptor is a critical determinant in receptor coupling to phosphoinositide phospholipase C- activating G proteins./biol Chem. 1996;271:23l Brightness Alters Heidelberg Retinal Flowmeter Measurements in an In Vitro Model Alexander C. Tsang, 1 Alon Harris, 12 Larry Kagemann, 1 Hak Sung Chung, 1 Bryan M. Snook, 1 and Hanna J. Garzozi 1 PURPOSE. The Heidelberg Retinal Flowmeter (HRF), a laser Doppler flowmetry device, has captured interest as a research and clinical tool for measurement of ocular blood flow. Concerns remain about the range and accuracy of the values that it reports. METHODS. An in vitro blood-flow model was constructed to provide well-controlled laminar flow through a glass capillary for assessment by HRF. A change in material behind the glass capillary was used to simulate changing brightness conditions between eyes. RESULTS. Velocities reported by the HRF correlated linearly to true velocities below 8.8 mm/sec. Beyond 8.8 mm/sec, HRF readings fluctuated randomly. True velocity and HRF reported velocities were highly correlated, with r = (P < 0.001) from 0.0 mm/sec to 2.7 mm/sec mean velocity using a light background, and r = (P < 0.001) from 2.7 mm/sec to 8.8 mm/sec using a darker background. How- From the Departments of 'Ophthalmology and 2 Physiology and Biophysics, Indiana University School of Medicine, Indianapolis. Supported in part by Grant EY (AH) from the National Institutes of Health, Bethesda, Maryland; and by an unrestricted grant from Research to Prevent Blindness, New York, New York. AH is a recipient of the William and Mary Greve Award from Research to Prevent Blindness. Submitted for publication June 19, 1998; revised October 5, 1998; accepted October 30, Proprietary interest category: N. Reprint requests: Alon Harris, Department of Ophthalmology, Indiana University School of Medicine, 702 Rotary Circle, Indianapolis, IN ever, a large change in the ^-intercept occurred in the calibration curve with the background change. CONCLUSIONS. The HRF may report velocities inaccurately because of varying brightness in the fundus. In the present experiment, a darker background produced an overreporting of velocities. An offset, possibly introduced by a noise correction routine, apparently contributed to the inaccuracies of the HRF measurements. Such offsets vary with local and global brightness. Therefore, HRF measurements may be error prone when comparing eyes. When used to track perfusion in a single eye over time, meaningful comparison may be possible if meticulous care is taken to align vessels and intensity controls to achieve a similar level of noise correction between measurements. (Invest Ophthalmol Vis Sci. 1999;40: ) The Heidelberg Retinal Flowmeter (HRF, Heidelberg Engineering, Germany) is a noninvasive confocal scanning laser imaging device marketed for the mapping of flow magnitudes in the human fundus. 1 It has several unique characteristics. 2 ' 3 Quantification of retinal blood flow is accomplished through a series of point measurements, each with a measurement resolution of approximately 10 X 10 /xm on the retinal plane, and a field depth of 400 /xm. An average clinical measurement takes approximately 5 minutes, including postmeasurement processing. Users of this technology have published various studies measuring the effects of disease and medications on blood flow in the fundus; 4 " 6 however, many questions on the validity of HRF still remain. One concerns the arbitrary unit in the HRF report. Despite previous experiments, 7 ' 8 physical units have not been successfully correlated to the arbitrary units. Furthermore, the functional range for these units, and the normal values, are yet to be determined. Can the HRF reliably differentiate glaucomatous disease processes from healthy ones based on these values? Are these values dependent on flow or other optical factors such as pallor? Our laboratory has used HRF to measure changes in blood flow induced by hypercapnia and hyperoxia. Were these changes representative of true changes in volumetric retinal blood flow? Based on these questions, this study had two specific purposes: to determine the range of linear response, or func-

2 796 Reports IOVS, March 1999, Vol. 40, No. 3 tional range, of the HRF in a highly controlled environment and to identify the effect of altered background brightness on HRF readings. METHODS A flow model was constructed using a horizontal heparinized glass pipette of 1.34-mm inner diameter. Bovine blood preserved and diluted with 3.2% sodium citrate solution (volume ratio, approximately 1:10) was fed by gravity from a large surface reservoir (80 cm 2 ), thereby maintaining a constant hydraulic head during each experimental trial. Flow was determined by weighing the mass of blood collected after each timed trial, which ranged from 240 to 1200 seconds. Flow velocity varied from 0 mm/sec to 30 mm/sec. Images of the flow were taken by HRF without optical magnification, and the resulting parameters at each velocity were averaged from 16 squares surrounding a fixed midline point of the pipette, with each square being 10 X 10 pixels. These squares overlap for the majority, shifting four pixels in each direction combined. Figure 1 shows a photograph of the set-up and the HRF display. The experiment was conducted in two series. In the first, a velocity range from 0.3 mm/sec to 2.67 mm/sec was obtained, and a background of white paper was used to create a bright background. The second series of trials ran with mean velocities from 2.71 mm/sec to 30 mm/sec. The background was altered in series two by replacing the white paper background with a less reflective white gauze. The change in background simulates conditions in which background and/or total image brightness is changed. Pixel brightness, labeled M(DQ in HRF reports, was then tallied for each series to complete the analysis. RESULTS In both series of trials, HRF measurements were linearly correlated to the true velocity of blood within the tube. The upper limit of linearity occurred when blood in the tube was moving at 8.83 mm/sec on average. The change of background scatter material between the two series induced a large change in the HRF velocity readings. The HRF velocity measurement was plotted against measured mean velocity (Fig. 2). It can be seen that the data were continuous within each operating condition, but both the slope and offset of the calibration curve were altered by the change in background conditions. There were high linearity and correlation within each series. The regression equation for series 1 (0.0 mm/sec to 2.67 mm/sec) was VhrC = f ^ Ktmc 40 arbitrary units (AU): 0 = black and 255 = white. The eye image had an second peak near 90 AU. Thirty percent of the pixels in the series 1 image were saturated: DC = 255. Each of these brightness distributions invoked a unique noise-correction curve within the HRF software. In Figure 4, the noise correction equations are shown. (r = 0.967; P< 0.001) and for series 2 (2.71 mm/sec to 8.83 mm/sec) was VM= FIGURE 1. A large surface area reservoir fed citrate-treated bovine blood through intravenous tubing to a heparinized pipette of 1.34 mm diameter, then along a glass slide (for surface tension control), and finally to a collection container. Inset is the HRF screen image of the pipette showing a square of examination at the midline of the pipette. This square was moved 4 pixels in each direction, resulting in 16 readings per velocity measurement, which were averaged before analysis. (r = 0.900; P< 0.001) To describe better the brightness of the paper and gauze compared with an image of an eye, the histogram in Figure 3 displays the distribution of brightness levels for series 1 with paper background, series 2 with cotton gauze background, and a healthy eye. All three distributions had a common peak near DISCUSSION There was a large change in the zero offset (Fig. 2, ^-intercepts) when the background was changed. The distribution of pixel brightness impacted significantly on the magnitude of HRF estimated noise (Fig. 4). The HRF noise estimation and correction algorithm, conveyed in correspondence with Heidelberg Engineering, is outlined next. Their noise estimation is based on the assumption that brighter images have proportionally more noise than darker images. In addition, it is dependent on

3 IOVS, March 1999, Vol. 40, No. 3 Reports ra 5 I Series 2 trials True mean velocity (mm/s) FIGURE 2. True mean velocity in the pipette is plotted against the velocity measurement of the HRF. A discontinuity in the data is clearly visible between the two trials. A plateau in velocity occurred at 8.83 mm/sec. Gradients of best-fit lines are 0.70 and 0.57 for series 1 and 2, respectively. A significant change in the ^-intercept occurred between the series, a result most probably of noise correction errors. the upper frequency band ( Hz) of the power spectra. The data processing is accomplished in two stages. The stage 1 computation involves noise estimation and correction for the entire image. It assumes and calculates the noise at every pixel as the raw spectral power in the 1500 Hz to 2000 Hz range. Then it correlates this assumed noise, N xy, to the intensity at all pixels, I xv, and computes a regression equation of the form (Fig. 4): intensity, I xy. The power spectrum is then corrected across all frequencies at ever}' pixel: In the stage 2 computation, values are reported at each pixel or selected square (averages): Flow is calculated from the moment of the corrected power spectrum weighted by intensity, synonymous with brightness in this context: NP{I) = ayjl+b. Next, it estimates for a second time the noise at each pixel, NP XJ,, from the regression equation above, based on the local where / is the frequency Brightness, M(DC), Interval l+jper Unfits in arbitrary units FIGURE 3. This histogram shows the distribution of pixel brightness, M(DC), from three images: an eye image, a pipette with paper background (series 1), and a pipette with a gauze background (series 2). Approximately 30% of the paper background image pixels had an M(DC) value of 255 arbitrary units (AU). Most pixels have brightness between 30 AU and 90 AU.

4 798 Reports IOVS, March 1999, Vol. 40, No. 3 eyaoorr gauzeoorr papercar Intensity of pxeis in arbitrary irrts (au) FIGURE 4. In the Heidelberg software, noise is estimated always to fit a distribution as pictured here. In the three images described in Figure 3, the resultant noise profiles are described by NP(J) = a\jl + b. Brightness of a pixel determines how much is subtracted from its power spectrum, but the noise profile is changed every time with a new image. Volume is calculated from the area of the corrected power spectrum weighted by intensity: Velocity is calculated as Volume xy =. Velocity = Flow, Volume xv To our knowledge, this algorithm has not been published or peer verified for flow measurement applications. It would produce higher flow readings for darker areas, because only a small correction would be subtracted from the raw scan spectrum. Because these corrections are derived globally and subtracted locally, the reported values are distorted between regions of the same image and among different images. This has four clinical consequences. First, when performing repeated examinations of a single normal eye, it is imperative to align the image perfectly from one examination to the next. Misalignment would change image content and illumination, both of which affect global brightness. Second, die illumination must be consistently set to the same level between images, by control of the sensitivity setting and camera-to-eye distance. Variations also induce changes in perceived noise. Third, different people would be expected to have different levels of fundus pigmentation and geometry. As a result, noise corrections would be different. In our opinion a simple intensity adjustment cannot compensate for the complex consequences of these noise corrections. Finally, and most damaging, the progressive disc pallor characteristic of glaucoma results in progressively darker images in longitudinal measurements, even under strict control. A flow change reported in these circumstances is likely to be confounded by the optical properties of the fundus. Our data are consistent with those in previous studies of the HRF. 7 ' 8 Others found a linear relationship only between 0.1 mm/sec and 1.0 mm/sec. 7 Our experiment resolution did not afford analysis below 0.1 mm/sec, where our only data points had zero flow velocity, but it agreed with previous assessments of linear relationship. Other ophthalmic laser Doppler devices display the same characteristic linear range ending at a plateau velocity, although at a level one order of magnitude higher than the HRF. 9 Based on a regression gradient of 0.7 AU per mm/sec and an 8.8-mm/sec velocity range, it can be deduced that the HRF can meaningfully report velocities in at a level of approximately 6 AU. Arbitrary units of velocities in the teens may be aberrations of the noise correction in addition to the effects of Doppler shifts. A similar argument would predict a level of approximately 2000 AU for flow. Overreporting in flow and velocity readings by a pure y-axis shift in the calibration curve has been observed previously. 7 We caution against universally applying the plateau velocity of 8.8 mm/sec as a clinical maximum. The sampling rate of 4000 Hz for the HRF would have predicted a plateau velocity of 0.78 mm/sec, compared with the 8.8 mm/sec in this study and the approximately 5 mm/sec in others. 8 The apparent extension of capability observed in our experiment probably depends on two factors: the penetration depth of the infrared laser in blood and the angle of incidence and scatter. The physical properties of transmission of light through blood are well known. 10 The intensity of HRF laser light as it passes through whole blood can be described as where / is the local intensity, I o is the initial intensity of the incident beam, 4.76 is the attenuation coefficient, and x is the distance of blood traveled through by the laser beam in milli-

5 IOVS, March 1999, Vol. 40, No. 3 Reports 799 meters. We calculated that 50% of the illuminating beam intensity had been absorbed or scattered by a depth of 150 /am, and 85% of the illumination intensity had been absorbed or scattered at 400 jam, the back of the focal plane. Because light must travel in and out of the blood column, the signal from the flow at a depth of 400 jam would be only 2.25% of the strength of the surface signal. We can assume that the signal majority originated from the superficial 75 jam (at least 50% of maximum signal). Further, the velocity profile for laminar flow in this capillary can be described as (R - V r =2V mc.j 1 - where r is the radial distance from the capillary wall into the flow column and r = 0.67 mm is the capillary radius. Approximately 70% of the incident beam is absorbed or scattered when light travels 130 /am into the tube. At a mean velocity of 8 mm/sec, the blood in the outer 130 jam of the capillary tube is moving at speeds between 0 mm/sec and 6.18 mm/sec. Because of limited penetration, we measured this slower moving blood and not the faster moving blood in the center. However, 6.18 mm/sec is still above the theoretic limit of 0.78 mm/sec for the HRF. This discrepancy may be in part because of the incidence angle. The debate on angle correction is unresolved. In our experiment, correcting for the incidence angle would result in sensed velocities that are only 8.72% of the mean. For the observed V mean of 8.5 mm/sec, therefore, the calculated velocity limit of the HRF is 0.54 mm/sec, or approximately 70% of what was expected because of sampling rate limitations. It is important to note that the above logic does not necessarily apply to the fundus or optic nerve head, where vessels are approximately 100 jam in diameter, where scatter from perfused tissue alters illumination directions, and where blood flow is no longer Newtonian. For comparison, the attenuation coefficient of perfused human dermal tissue in vivo is closer to 1.28, or one fourth of whole blood. Thus the true absorption characteristics of retinal tissue vary depending on the density of the vasculature. The random and competing effects of media scattering, in addition to angle of incidence may explain the remainder of the discrepancies. The effects of these variables were not measured this study. Because of its sensitivity to moving blood flow, the HRF is successful in producing dramatic images of vasculature without intravessel contrast. However, it produces this flowweighted vasculature map while misrepresenting the numerical values it is supposed to report. The ideal flow mapping system should discern and quantify velocity changes apart from brightness changes for three reasons: First, velocity changes alone may be the early sign of a presymptomatic disease process. Second, physiological changes or drugs may alter perfusion and cause local anatomic changes and therefore brightness changes (e.g., vasodilatation), but perfusion may also be altered with pure velocity changes (e.g., upstream pressure regulation). Third, brightness is dependent on the image acquisition process, which introduces additional variations, as we showed. CONCLUSIONS We discovered that HRF measurements can be drastically altered by the brightness of an image. Our results suggest that the HRF noise reduction algorithm may account for much of the variations and false flow measurements produced by the instrument. These findings present a new dilemma for comparative studies vising HRF. Intersubject comparison may be unreliable with this software version, because eye image brightness depends on individual anatomy. To track a single eye over time, great care must be taken to minimize variability of disc location within the image. HRF sensitivity should be held constant between images during longitudinal follow-up of individual eyes. Changes in the scattering and reflective properties of the fundus such as edema or disc pallor may introduce significant changes in flow values as a result of noise correction. Because of optical characteristics of blood and human tissue, the HRF arbitrary units may relate to actual flow velocities by a complex relationship involving angle of incidence, blood cell density, and vessel diameter. The linearity and plateau of the HRF response curve are consistent with previous experiences with the technology. It is clear that the HRF is velocity sensitive, and a linear response can be obtained with strict control over factors that affect the noise correction. In its present form, the HRF is not a reliable flow measurement device. A long-term solution to enhance die usability of the HRF should begin with a re-examination of the noise estimation and correction algorithm. References 1. Michelson G, Schmauss B. Two-dimensional mapping of the perfusion of the retina and optical nerve head. Br J Ophthalmol. 1995;79: Riva CE, Cransoun SE, Grunwald JE, Petrig BL. Choroidal blood flow in the foveal region of the human ocular fundus. Invest Ophthalmol Vis Sd. 1994;35: Riva CE, Petrig BL, Grunwald JE. Near infrared retinal laser Doppler flowmetry. Lasers Ophthalmol. 1987;1: Park K-H, Choi SY, Park KH, Lee J. Evaluation of parafoveal blood flow in branch retinal vein occlusion (BRVO) patients using Heidelberg Retina Flowmeter (HRF) [ARVO Abstract]. Invest Ophthalmol Vis Sci. 1997;34(4):S774. Abstract nr Ness T, Muller-Velten R, Funk J. No changes in ocular blood flow of the retina after topically applied Timolol: Heidelberg Retina Flowmeter parameters in healthy volunteers [ARVO Abstracts]. Invest Ophthalmol Vis Sci. 1997;34(4):S777. Abstract nr Kagemann L, Harris A, Chung HS, Evans D, Buck S, Martin B. Heidelberg retinal flowmetry: factors affecting blood flow measurement. Br J Ophthalmol. 1998;82: Chauhan BC, Smith FM. Confocal scanning laser Doppler flowmetry: experiments in a model flow system. / Glaucoma. 1997;6: Van Heuven Waj, Kiel JW, Elliot WR, Harrison JM, Sponsel WE. Evaluation of the Heidelberg Retina Flowmeter [ARVO Abstract]. Invest Ophthalmol Vis Sci. 1996;37(3):S967. Abstract nr Obeid AN, Barnett NJ, Dougherty G, Ward G. A critical review of laser Doppler flowmetry./med Eng Tech. 1990;l4: Cui WJ, Ostrander LE, Lee BY. In vivo reflectance of blood and tissue as a function of light wavelength. IEEE Trans Biomed Eng. 1990;37: