Imaging the Cerebral Blood Flow with Enhanced Laser Speckle Contrast Analysis (eLASCA) By Monotonic Point Transformation Peng Miao, Minheng Li, Hugues Fontenelle, Anastasios Bezerianos, Member, IEEE, Yihong Qiu, Yisheng Zhu, Senior Member, IEEE, and Shanbao Tong*, Member, IEEE

Abstract Laser speckle contrast analysis (LASCA) has been demonstrated as a full-field method for imaging the cerebral blood flow (CBF). However, conventional LASCA is limited to extremely low dynamic range because of the ambient background field, dark current and anomalies in the circuits of CCD camera, which makes it difficult to analyze the spatiotemporal variabilities in CBF. In this study, we proposed an enhanced laser speckle contrast analysis (eLASCA) method to improve the dynamic range of LASCA based on monotonic point transformation (MPT). In investigating the influence of moderate hypothermia (32 ± 0.5◦ C) on capillary CBF change, eLASCA presented much more significant (189%) decrease of CBF under hypothermia than LASCA (137%). Statistically, eLASCA resulted in a higher confidence degree(p = 0.009) of CBF change after the rewarming than LASCA (p = 0.013). In addition, eLASCA greatly improves the CBF visualization, which is very helpful in demonstrating the details of CBF change.

Index Terms laser speckle contrast analysis (LASCA), cerebral blood flow (CBF), dynamic range enhancement, monotonic point transformation (MPT), hypothermia

P. Miao, M. Li, Y. Qiu, and Y. Zhu are with the Department of Biomedical Engineering, Shanghai Jiao Tong University, Shanghai, 200240 P.R. China. S. Tong is with Med-X Research Institute, Shanghai Jiao Tong University, Shanghai, 200230 P.R. China. H. Fontenelle and A. Bezerianos are with the Departement of Medical Physics, University of Patras, Patras, Greece. *Please send the correspondence to Dr. Shanbao Tong, e-mail: [email protected]. April 2, 2008

DRAFT

MANUSCRIPT SUBMITTED TO IEEE TRANSACTIONS ON BIOMEDICAL ENGINEERING

1

Imaging the Cerebral Blood Flow with Enhanced Laser Speckle Contrast Analysis (eLASCA) By Monotonic Point Transformation I. I NTRODUCTION Laser speckle contrast analysis (LASCA) for imaging the cerebral blood flow (CBF) has been rapidly developed since Dunn and colleges firstly presented it in 2001[1][2][3]. LASCA has been used in the studies of CBF changes under peripheral electrical stimulation[4], single vibrissa stimulation[5][6], focal cerebral ischemia[3], hypotension[7], dorsal skin fold[8] and tumor[9]. LASCA is based on the fact that when coherent laser light illuminates a surface with moving scattering particles, e.g. blood cells in vessels, the temporal and spatial intensity flunctuations of the speckle patterns in the image from the reflected light are averaged out[10]. Practically, relative velocity of CBF can be estimated by the spatial[2] or temporal[11] contrast analysis, i.e. sLASCA or tLASCA. Mathematically, the contrast is defined as the ratio of the standard deviation to the ensemble average of the speckle intensities [2]. K=

σs hIi

(1)

In the study of blood flow, K is linked to the velocity v by the following equation [1] K2 =

1 τc (1 − e−2T /τc ) = (1 − e−(2ako T )v ) 2T (2ako T )v

(2)

where T is the exposure time of the camera[12], ko is the light wavenumber, and the factor a depends on the Lorentzian width and scattering properties of the tissue[13]. Eq.(2) implies that the square of contrast, i.e. K 2 , is approximately proportional to 1/v when K 2 is small. Therefore, we use K 2 instead of K to study the CBF changes hereafter. Practically, the temporal laser speckle contrast analysis (tLASCA)[11] is used to improve the spatial resolution with more capillary details. K2 =

σt2 hIi2

(3)

According to Zakharov et al.[14], Eq.(3) assumes that the deterministic autocorrelation function of the CCD exposure window is 1 when the exposure time T grows sufficiently large[15]. Wang et al. demonstrated that this assumption was correct, even for fast CCD cameras[16]. April 2, 2008

DRAFT

MANUSCRIPT SUBMITTED TO IEEE TRANSACTIONS ON BIOMEDICAL ENGINEERING

2

The theoretical value of K 2 is within the range of [0, 1]. In practice, however, the dynamic range of K 2 is extremely limited due to the ambient background field, the effect of dark current and other anomalies in the CCD camera[17][18][19]. K 2 values in most literatures were within [0, 0.1], e.g. K 2 < 0.01 in Dunn’s work[3], K 2 < 0.04 in Cheng’s work[11] and K 2 < 0.09 in Yuan’s work[20]. Fig.2 (a) shows a typical K 2 in this study with all values falling into [0, 0.048] and 99% within [0, 0.02] (see the probability density function (p.d.f.) in Fig.3). By far, there is no effective method to eliminate the influences from ambient background field, dark current and other noise[17]. Besides, such a limited dynamic range not only influences the comparison of CBF changes, but also makes it difficult to analyze the transient spatiotemporal changes of blood flow in microvascular level. Furthermore, a limited and low dynamic range of contrast leads to a extremely large range of 1/K 2 (∝ v ), such that the velocity visualization would be problematic by traditional image processing methods. In this paper, we propose an enhanced laser speckle contrast analysis (eLASCA) by monotonic point transformation (MPT) to solve the above problems. eLASCA improves the dynamic range of K 2 and keeps the variabilities of contrast values. Besides, eLASCA method is fully adaptive and low computation load, which is suitable for a real-time or on-line processing. II. M ETHODS AND M ATERIALS A. Enhanced Laser Speckle Analysis (eLASCA) Suppose L trials, T frames each, of laser speckle images (M × N pixels) were acquired in a single experiment, i.e. Il (m, n, t), (m = 1, · · · , M ; n = 1, · · · , N ; t = 1, · · · , T ; l = 1, · · · , L). The corresponding contrast matrix K 2 (m, n, l) can be obtained by Eq.(3). In order to improve the dynamic range, we reshape three dimensional K 2 into one-dimensional random variable f (i), (i = 1, · · · , M × N × L), by Eq.(4): f (m + (n − 1) × M + (l − 1) × M × N ) = K 2 (m, n, l)

(4)

The profile of the p.d.f. of f , i.e. p(f ), is very sharp in a small range close to zero (Fig.3). To improve the dynamic range, f is transformed into fe satisfying p(fe ) ≡ 1 by MTP theory[21]. The transform does not change the validity of f according to the monotonicity of the transform: Z

fe

0

Z

p(fe ) dfe =

0

f

p(f ) df

(5)

where both f and fe are limited in [0,1].

April 2, 2008

DRAFT

MANUSCRIPT SUBMITTED TO IEEE TRANSACTIONS ON BIOMEDICAL ENGINEERING

3

Since p(fe ) ≡ 1, Eq.(5) can be easily solved: Z

fe =

f

0

(6)

p(f ) df

According to Eq.(2), K 2 is approximately proportional to 1/v , therefore, f can be represented as: f=

b v

(7)

p(f ) df

(8)

where b is a constant. Combining Eq.(6) and Eq.(7), we have: Z

fe =

0

b v

Since the contrast is always non-negative, the relation between v and fe can be numerically induced from Eq.(8) with the cumulative density function (c.d.f.) C(f ) =

Rf

−∞ p(f ) df [22]:

b b fe = C( ) − C(0) = C( ) v v

(9)

Clearly, higher velocity v results in lower f , and then leads to lower fe after MPT. In practice, Eq.(6) can be approximately estimated with the percentage of contrast value less than or equal to f , i.e. N umf , in K 2 : fe =

N umf × 100% M ×N ×L

(10)

In summary, the procedures of eLASCA are: 1) Reshape 3 dimensional contrast K 2 to 1 dimensional vector f by Eq.(4). 2) Sort the vector f by ascending order and say it as g . 3) Compute fe based on g by Eq.(10). 4) Reshape the vector fe back to 3 dimensional Ke2 by Eq.(4). B. Animal Preparation Seven male Sprague-Dawley rats (300 ± 50g) were used to obtain the laser speckle images under experimental hypothermia protocols approved by the Committee for Animal Care and Use of Shanghai Jiao Tong University. The animals were anaesthetized with pentobarbital (80mg/kg, IP) and mounted in a stereotaxic frame (Benchmark DeluxeT M ; MyNeurolab.com, St. Louis, MO). In the surgery, a midline incision was made to expose the skull. A window (4.5 × 6.5 mm) overlying the right barrel cortex (5.5 mm lateral, 2.5 mm posterior to the bregma) was thinned with a high speed dental drill (Stoelting, U.S.A) equipped with 1.6mm drill burr (Dentsply, Switzerland). The thinned area was filled with glycerine to

April 2, 2008

DRAFT

MANUSCRIPT SUBMITTED TO IEEE TRANSACTIONS ON BIOMEDICAL ENGINEERING

4

reduce the glaring reflection. A feedback-controlled heating pad (DC Temperature Control Module, FHC Inc., Bowdoin, ME) was used to maintain the body temperature. A semiconductor laser diode (635 nm) (20mW KL5650, Forward Co.,Ltd., Shanghai, China) was used to illuminate the thinned window. A monochrome 12-bit CCD camera (Pixelfly QE, Cooke, USA) with resolution of 1024 × 1392 pixels was positioned over the thinned skull and focused on the cortical surface by a trinocular stereo zoom boom microscope (XYH-05, Shanghai Optical Instrument Factory, Shanghai, China). A 2 × 2 hardware camera binning was implemented such that 512×696 pixel laser speckle images were output to computer for CBF analysis. Exposure time of the camera was set to 5ms[20] and the imaging rate was 23 fps throughout the experiments. The experimental setup is shown in Fig.1. C. Data Recording During the experiment, the temperature of the rat was monitored with a rectal probe and maintained by the feedback-controlled heating lamp and pad. 20 min after the surgery preparation, the first trial of laser speckle images (l = 1, T = 200) were acquired as the baseline (37◦ C). Then, whole-body hypothermia was induced by 30 min of surface cooling with alcohol bathing. After 20 min maintaining the rectal temperature at 32 ± 0.5◦ C, another 200 frames of laser speckle images were captured as the hypothermia trial (l = 2). The 30 min rewarming procedure started after 40 min of hypothermia until the rectal temperature reached 37 ± 0.5◦ C. Then after another 20 min at 37◦ C, the last trial (l = 3) laser speckle images (T = 200) were recorded. D. Data Analysis For each rat, we got three trials (l = 1, 2, 3) of laser speckle images (T = 200 each). After processing each trial with LASCA (Eq.(3)), a 512 × 696 × 3 contrast matrix K 2 (m, n, l) were obtained for baseline (l = 1), hypothermia (l = 2), and post-rewarming(l = 3) trial, respectively. Then, K 2 of each rat was processed by eLASCA method to improve the dynamic range, i.e. Ke2 . III. R ESULTS A. Automatic Visualization of the Contrast Data To visualize the contrast data, we need convert K 2 into a number of gray levels. Fig.2(a) showed the typical K 2 (m, n, l) in baseline trial (l = 1). Because the majority of contrast data fall into an extremely small dynamic range [0, 0.02], the image was too dark to show the details. In order to properly visualize the CBF information, most literatures simply converted the main part of the data April 2, 2008

DRAFT

MANUSCRIPT SUBMITTED TO IEEE TRANSACTIONS ON BIOMEDICAL ENGINEERING

5

to gray levels instead of the full range, e.g. K 2 ∈ [0, 0.02] as shown in Fig.2(b). Such a visualization processing not only needs manual intervention, but also loses the information of some data. Fig.2(c) illustrated the automatic visualization by conventional contrast limited adaptive histogram equalization (CLAHE) [23][24]. Compared with Fig.2 (a,b,c), eLASCA presented more vascular CBF details (Fig.2 (d)), particularly in the capillaries. Furthermore, the traditional method like CLAHE is not able to deal with the velocity visualization because an extremely limited low dynamic range (e.g. [0, 0.048] in Fig.2(a)) result in a unlimited velocity range (e.g. [20, ∞]) while eLASCA visualizes the 1/K 2 perfectly by reducing the range of 1/K 2 to [0, 1]. Fig.4 (a, b, c) showed the velocity changes of CBF under baseline, hypothermia and post-rewarming conditions. B. Influences of Hypothermia on Capillary CBF Fig.5 showed the LASCA (K 2 ) for baseline (Fig.5(a)), hypothermia (Fig.5(b)) and post-rewarming (Fig.5(c)) trials in one experiment, respectively. As a comparison, the corresponding eLASCA (Ke2 ) were also illustrated (Fig.5(d)-(f)). There were clear CBF differences between hypothermia and normothermia trials. In capillary area, eLASCA seemed to present more evident decrease of CBF under hypothermia. To quantitatively analyze the CBF changes in capillary level, we did cortical arteries, veins(Fig.5 (g)) and capillaries(Fig.5 (h)) segmentation by Otsu multi-threshold method[25] in ITK software[26](bin=256 and threshold number=4). Considering the variation of CBF in baseline trial, we analyzed the relative CBF change. Fig.5 (i) showed the changes of capillary CBF from seven rats in hypothermia and post-rewarming trials by LASCA and eLASCA respectively. Under hypothermia, eLASCA showed much greater decrease (189%) of capillary CBF than LASCA (137%). While in post-rewarming trial, capillaries have more significant recovery by eLASCA (151%) than by LASCA (119%). Besides, the statistic analysis by paired t-test of CBF recovery indicated that eLASCA (p = 0.009) provided higher confidence level than LASCA (p = 0.013). IV. D ISCUSSION eLASCA not only provides an automatic visualization algorithm for the CBF change, but also keeps the CBF variability after the transform. Fig.8 plots the contrast values at the dashed line in Fig.2 by LASCA, Histogram Equalization and eLASCA, respectively. Compared with the conventional LASCA, both Histogram Equalization and eLASCA greatly improve the dynamic range of the contrast data. However, Histogram Equalization lost the data variabilities when the fluctuations of CBF are within an April 2, 2008

DRAFT

MANUSCRIPT SUBMITTED TO IEEE TRANSACTIONS ON BIOMEDICAL ENGINEERING

6

extremely small range, as highlighted in the dashed windows in Fig.8. The reason is that the histogram equalization processing is based on the image histogram, which leads the loss of variations in data when applied to a continuous random variable (Fig.6 (b)). Keeping the data variability helps to segment the vessels. For example, cortical capillary bleeding in surgery sometimes is not avoidable,(see Fig.7 (a)), which may result in artifacts in vessel segmentation (Fig.7 (b)). Fig.7 (c) showed much more vascular details by the same method in Fig.7 (b) after eLASCA transform. V.

CONCLUSIONS

In this paper, we proposed MPT based eLASCA method to improve the dynamic range of contrast, while maintaining the variability of data. eLASCA is fast and does not require supervision as a visualization tool. Compared with the conventional LASCA technique, eLASCA presents much better performances in visualizing and analyzing the CBF variability information. ACKNOWLEDGMENT This work is partly supported by the grant of Shanghai Pujiang Program (06PJ14055) and by a research grant of the University of Patras, Karatheodoris (2004-B411): Functional Laser Speckle Imaging (fLSI). S.T. is also Supported by Program for New Century Excellent Talents in University.

April 2, 2008

DRAFT

MANUSCRIPT SUBMITTED TO IEEE TRANSACTIONS ON BIOMEDICAL ENGINEERING

7

R EFERENCES [1] A. Fercher and J. Briers, “Flow visualization by means of single-exposure speckle photography,” Optics Communications, vol. 37, no. 5, pp. 326–330, 1981. [2] J. Briers and S. Webster, “Laser Speckle Contrast Analysis (LASCA): A Nonscanning, Full-Field Technique for Monitoring Capillary Blood Flow,” Journal of Biomedical Optics, vol. 1, p. 174, 1996. [3] A. Dunn, H. Bolay, M. Moskowitz, and D. Boas, “Dynamic Imaging of Cerebral Blood Flow Using Laser Speckle,” Journal of Cerebral Blood Flow & Metabolism, vol. 21, pp. 195–201, 2001. [4] T. Durduran, M. Burnett, G. Yu, C. Zhou, D. Furuya, A. Yodh, J. Detre, and J. Greenberg, “Spatiotemporal Quantification of Cerebral Blood Flow During Functional Activation in Rat Somatosensory Cortex Using Laser-Speckle Flowmetry,” Journal of Cerebral Blood Flow & Metabolism, vol. 24, pp. 518–525, 2004. [5] B. Weber, C. Burger, M. Wyss, G. von Schulthess, F. Scheffold, and A. Buck, “Optical imaging of the spatiotemporal dynamics of cerebral blood flow and oxidative metabolism in the rat barrel cortex,” European Journal of Neuroscience, vol. 20, no. 10, pp. 2664–2670, 2004. [6] W. Lau, S. Tong, and N. Thakor, “Spatiotemporal characteristics of low-frequency functional activation measured by laser speckle imaging,” Neural Systems and Rehabilitation Engineering, IEEE Transactions on [see also IEEE Trans. on Rehabilitation Engineering], vol. 13, no. 2, pp. 179–185, 2005. [7] A. Kharlamov, B. Brown, K. Easley, and S. Jones, “Heterogeneous response of cerebral blood flow to hypotension demonstrated by laser speckle imaging flowmetry in rats,” Neuroscience Letters, vol. 368, no. 2, pp. 151–156, 2004. [8] B. Choi, N. Kang, and J. Nelson, “Laser speckle imaging for monitoring blood flow dynamics in the in vivo rodent dorsal skin fold model,” Microvascular Research, vol. 68, no. 2, pp. 143–146, 2004. [9] D. Zhu, W. Lu, Y. Weng, H. Cui, and Q. Luo, “Monitoring thermal-induced changes in tumor blood flow and microvessels with laser speckle contrast imaging,” Applied Optics, vol. 46, no. 10, pp. 1911–1917, 2007. [10] J. Briers, “Laser Doppler, speckle and related techniques for blood perfusion mapping and imaging,” Physiol. Meas, vol. 22, no. 4, 2001. [11] H. Cheng, Q. Luo, S. Zeng, S. Chen, J. Cen, and H. Gong, “Modified laser speckle imaging method with improved spatial resolution,” Journal of Biomedical Optics, vol. 8, p. 559, 2003. [12] J. Goodman, “Some effects of target-induced scintillation on optical radar performance,” Proceedings of the IEEE, vol. 53, no. 11, pp. 1688–1700, 1965. [13] R. Bonner and R. Nossal, “Model for laser Doppler measurements of blood flow in tissue,” Appl. Opt, vol. 20, no. 12, pp. 2097–2107, 1981. [14] P. Zakharov, A. Voelker, A. Buck, B. Weber, and F. Scheffold, “Quantitative modeling of laser speckle imaging,” Arxiv preprint cond-mat/0606030, 2006. [15] A. V¨olker, P. Zakharov, B. Weber, F. Buck, and F. Scheffold, “Laser speckle imaging with an active noise reduction scheme,” Optics Express, vol. 13, no. 24, pp. 9782–9787, 2005. [16] Z. Wang, S. Hughes, S. Dayasundara, and R. Menon, “Theoretical and experimental optimization of laser speckle contrast imaging for high specificity to brain microcirculation,” Journal of Cerebral Blood Flow & Metabolism, vol. 27, pp. 258–269, 2007. [17] G. Richards and J. Briers, “Capillary-blood-flow monitoring using laser speckle contrast analysis (LASCA): improving the dynamic range,” Proceedings of SPIE, vol. 2981, p. 160, 1997.

April 2, 2008

DRAFT

MANUSCRIPT SUBMITTED TO IEEE TRANSACTIONS ON BIOMEDICAL ENGINEERING

8

[18] C. Roundy, G. Slobodzian, K. Jensen, and D. Ririe, “Digital imaging produces fast and accurate beam diagnosis,” Laser Focus World, vol. 29, no. 10, pp. 117–125, 1993. [19] S. Webster and J. Briers, “Time-integrated speckle for the examination of movement in biological systems,” Proceedings of SPIE, vol. 2132, pp. 444–452, 1994. [20] S. Yuan, A. Devor, D. Boas, and A. Dunn, “Determination of optimal exposure time for imaging of blood flow changes with laser speckle contrast imaging,” Applied Optics, vol. 44, no. 10, pp. 1823–1830, 2005. [21] W. Pratt, Digital Image Processing: PIKS Inside. John Wiley & Sons, Inc. New York, NY, USA, 2001. [22] D. Cox and D. Oakes, Analysis of Survival Data. Chapman & Hall/CRC, 1984. [23] K. Zuiderveld, “Contrast limited adaptive histogram equalization,” Academic Press Graphics Gems Series, pp. 474–485, 1994. [24] A. Reza, “Realization of the Contrast Limited Adaptive Histogram Equalization (CLAHE) for Real-Time Image Enhancement,” The Journal of VLSI Signal Processing, vol. 38, no. 1, pp. 35–44, 2004. [25] P. Liao, T. Chen, and P. Chung, “A Fast Algorithm for Multilevel Thresholding,” Journal of Information Science and Engineering, vol. 17, no. 5, pp. 713–727, 2001. [26] L. Ibanez and W. Schroeder, “The ITK Software Guide: The Insight Segmentation and Registration Toolkit. Kitware,” Inc., Albany, NY, www. itk. org, vol. 1, 2003.

April 2, 2008

DRAFT

MANUSCRIPT SUBMITTED TO IEEE TRANSACTIONS ON BIOMEDICAL ENGINEERING

9

Figure Captions Fig.1: The setup of our experiments. Fig.2: Visualization of contrast data in this study. (a) showing the LASCA data directly. (b) showing the LASCA data in the range [0, 0.02] selected manually. (c) showing the LASCA data by CLAHE method automatically. (d) showing the eLASCA data directly. Fig.3: The probability density function (p.d.f.) and the cumulative density function (c.d.f.) of the LASCA data in Fig.2. Fig.4: The results of eLASCA method for visualizing the 1/K 2 data (∝ v ): baseline (a), hypothermia (b), and post-rewarming (c) trial. Fig.5: The analysis of 3 dimensional contrast data by eLASCA. The LASCA data matrix K 2 of one experiment: baseline (a), hypothermia (b), and post-rewarming (c) trial respectively. The corresponding results (d,e,f) of eLASCA. The segmentations for artery and vein (g) and capillary(h) by Otsu method.The relative CBF change (i). Fig.6: eLASCA maintains the data precision while histogram equalization can not. (a) The 170th raw of LASCA data. (b) The result of the histogram equalization method. (c) The 170th raw of eLASCA data. Fig.7: eLASCA improved the segmentation result of image with bleeding points. (a) the eLASCA contrast data. (b) Result based on LASCA data by Otsu method. (c) Result based on eLASCA data by Otsu method.

April 2, 2008

DRAFT

MANUSCRIPT SUBMITTED TO IEEE TRANSACTIONS ON BIOMEDICAL ENGINEERING

10

Fig. 1.

Figure

April 2, 2008

DRAFT

MANUSCRIPT SUBMITTED TO IEEE TRANSACTIONS ON BIOMEDICAL ENGINEERING

11

0.02

0.048

a

b

0.04 0.015 0.03 0.01 0.02 0.005

0.01 0 1

0

c

1

d

0.8

0.8

0.6

0.6

0.4

0.4

0.2

0.2

0

0

Fig. 2.

Figure

April 2, 2008

DRAFT

MANUSCRIPT SUBMITTED TO IEEE TRANSACTIONS ON BIOMEDICAL ENGINEERING

12

1 pdf cdf

p.d.f. and c.d.f.

0.8 0.6 0.4 0.2 0

0

0.005

0.01

0.015

0.02 0.025 0.03 Contrast Values (f)

0.035

0.04

0.045

Fig. 3.

Figure

April 2, 2008

DRAFT

MANUSCRIPT SUBMITTED TO IEEE TRANSACTIONS ON BIOMEDICAL ENGINEERING

a

b

13

c

Fig. 4.

Figure

April 2, 2008

DRAFT

MANUSCRIPT SUBMITTED TO IEEE TRANSACTIONS ON BIOMEDICAL ENGINEERING

14

0

a

b

c 0.006 0.012 0.018 0.024 0.03

0

d

e

f 0.2 0.4 0.6 0.8 1

g

h

250%

LASCA

200%

eLASCA

i

150% 100% 50% 0%

Hypothermia

Post−rewarming

Fig. 5.

Figure

April 2, 2008

DRAFT

MANUSCRIPT SUBMITTED TO IEEE TRANSACTIONS ON BIOMEDICAL ENGINEERING

15

LASCA

0.03 0.02 0.01 0

50

100

150

200

250

50

100

150

200

250

50

100 150 Pixel’s Column Position

200

250

Histogram Equalization

1 0.5 0

eLASCA

1 0.5 0

Fig. 6.

Figure

April 2, 2008

DRAFT

MANUSCRIPT SUBMITTED TO IEEE TRANSACTIONS ON BIOMEDICAL ENGINEERING

a

b

16

c

Fig. 7.

Figure

April 2, 2008

DRAFT