You are on page 1of 5

3rd National Conference on Signal Processing, Communications and VLSI Design (NCSCV ’11) 6th & 7th May 2011

High Resolution Cerebral Blood Flow (CBF) Image Enhancement by Registered Laser Speckle Contrast Analysis (rLASCA)
A. Kumaran* B. Elango**
* Senior Lecturer, Sri Sairam Engineering College, Chennai-44 manniappan@yahoo.com **Final year Student-M.E-Communication systems, Sri Sairam Engineering College, Chennai-44 beim.18@gmail.com

Abstract—Laser speckle imaging (LSI) has been widely used for in vivo detecting cerebral blood flow (CBF) under various physiological and pathological conditions. So far, nearly all literature on in vivo LSI does not consider the influence of disturbances due to respiration and/or heart beating of animals. In this paper, we analyze how such physiologic motions affect the spatial resolution of the conventional laser speckle contrast analysis (LASCA).We propose a registered laser speckle contrast analysis (rLASCA) method which first registers raw speckle images with a 3×3 convolution kernel, normalized correlation metric and cubic B-spline interpolator, and then constructs the contrast image for CBF. rLASCA not only significantly improves the distinguish ability of small vessels, but also efficiently suppresses the noises induced by respiration and/or heart beating. In an application of imaging the angiogenesis of rat’s brain tumour, rLASCA outperformed the conventional LASCA in providing a much higher resolution for new small vessels. As a processing method for LSI, rLASCA can be directly applied to other LSI experiments where the disturbances from different sources (like respiration, heart beating) exist. Index Terms—Cubic B-spline interpolation, laser speckle imaging (LSI), normalized correlation registration, registered laser speckle contrast analysis (rLASCA).

K2 = (σ/I) 2 α 1/v
K2 can be spatial or temporal. For each frame of the acquired speckle images, K s2 for each pixel is calculated based on the intensities in a small window (typically 7×7), and therefore loses the spatial resolution. Similarly, Kt2 at each pixel is estimated using the intensities of the pixel in continuously recorded raw speckle images. In practice, 50 or more frames are used for statistical efficiency. As a result, Kt2 retains the spatial resolution but temporal resolution is compromised. Like other optical imaging techniques, LSI is also susceptible to kinds of mechanical vibrations within the workspace. Such mechanical vibrations can be minimized by using vibration isolated optical platform and other precautions. In rodent in vivo experiments, the animals are always anesthetized and constrained in a stereotaxic frame to suppress movements. However, there are still some inevitable disturbances in raw speckle images due to respiration and/or heart beating of the animals. Such global disturbances could lead to the following problems in LASCA: 1) the loss of spatial resolution, the output of LASCA is going to be blurred, and thus it is more difficult to distinguish small vessels from tissue; 2) inaccurate estimation of contrast values, disturbances change the statistic property of the speckles and lead to inaccurate estimation of CBF by LASCA. In order to overcome these problems, we propose to register the raw speckle images before LASCA, or called registered LASCA (rLASCA) hereafter. Since the raw speckle images are usually too noisy to find a stable pattern for registration, we develop a novel automatic registration technique, using a 3X3 convolution kernel, a normalized correlation metric and a cubic Bspline interpolator, to register raw speckle images accurately. rLASCA is shown to enhance the small vessels and alleviate noises in both tissue area and vessels. As an application, we studied the angiogenesis in rat’s brain tumour by LSI, and showed how rLASCA improves the resolution of cortical vascular structure of the angiogenesis.

I.INTRODUCTION
Laser Speckle Imaging (LSI) has been widely used to study rat’s cerebral blood flow (CBF) under a variety of physiological and pathological conditions for a long time. When the tissue contains scattering moving particles, e.g., blood cells, sequentially recorded raw speckle are analyzed by laser speckle contrast analysis (LASCA) to obtain both CBF and vessel structure information. According to the theory of LASCA, the contrast value K2, defined as the square of ratio of standard deviation σ to mean intensity (I), is inversely proportional to velocity v.

II. PROPOSED WORK

Organized by Department of ECE, Anna University of Technology, Coimbatore

2. Communications and VLSI Design (NCSCV ’11) 6th & 7th May 2011 rLASCA involves two steps A) Registration of the raw speckle images and B) Performing LASCA based on registered image to analyze the raw speckle image for contrast value and thereby calculating velocity of blood flow. . continuous and bounded. . cubic B-spline interpolation prevents the effect of degeneration in linear interpolation. . In addition. The pre-processed images reveal the same vessel structure pattern which. After re-sampling. N). First the raw speckle image is pre-processed by using a convolution kernel to reveal the vessel structure pattern. and the last column) of Si is lost. . therefore. . because the random speckles indicate no stable pattern. Obtaining Rigid Translation Parameters based on Normalized Correlation Metric In this step. A) REGISTRATION OF RAW SPECKLE IMAGE Usually registration techniques are incapable of processing raw speckle images {Ii(x. . Then a rigid translation parameter is estimated for each pre-processed image.N). by a normalized correlation metric. so the rigid translation parameters (Δxi. β3 (d) = A 3X3 convolution kernel is implemented to preprocess each raw speckle image to obtain the spatial standard deviation at each pixel. .. 3. The cubic B-spline interpolator is used to calculate the intensity of non-grid position in a two dimensional piecewise polynomial way based on the cubic B-spline function which is smooth. it is used in this study because cubic B-spline interpolation utilizes the noninteger parameters which are preferable over the nearest neighbour interpolation. Si2 =(Ii2 ⊗ Kernel) . each image Si (i=2. a re-sampling interpolation is needed to register the raw speckle images {Ii(x.(Ii ⊗ Kernel)2 The size of image Si is equal to that of Ii. Since the convolution kernel size is 3X3. N) are obtained.3rd National Conference on Signal Processing. Re-Sampling Raw Speckle Images using Cubic B-Spline Interpolator The displacement due to disturbances need not conform to integral pixels. only the information at the four borders (the first row. Finally the raw speckle images are registered (re-sampled) by cubic b-spline interpolator based on their rigid translation metric to obtain registered raw speckle images. In this paper. 1. Where |d| is distance from interpolated point. Δyi) for each Si (i = 2. N). Metric = Organized by Department of ECE. Pre-processing using 3X3 Convolution Kernel There are various registration metrics but the metric based on normalized correlation metric is chosen because of its less computational cost and well defined sharp peaks in cost function. The optimized rigid translation parameter (Δxi. which could be used for registration. Therefore. y)} (i = 1. the registered raw speckle images Ii (i =1. Coimbatore . the first column. . Although the computational cost of B-spline interpolation is not very low. is to be registered to the first image S 1 in accordance with a proper rigid translation. the last row. the average loss of spatial resolution in this pre-processing step is very small and always less than one pixel. we propose a registration technique to register the raw speckle images. y)}(i = 1. can be used for registering with a minor loss of spatial resolution. . which is related to vessel structure information. The registration process of the raw speckle image can be shown as a block diagram. Δyi) always have sub pixel precision. is determined by maximizing the normalized correlation metric.. . N). Anna University of Technology. Therefore.

disturbances are Figure 4: Figure showing the registered image Organized by Department of ECE. The ratio of standard deviation to the mean intensity of that pixel is calculated 3.3rd National Conference on Signal Processing. Then this contrast value is applied to the central pixel in the block and this step is repeated for all the blocks. The next step is to alleviate these disturbances due to noise from heartbeat. III. RESULTS The main objective of this project is to enhance the raw speckle image so that the small vessels can also be seen clearly. These contrast values are then given false colour map or the velocity value. Coimbatore . 4. But this technique without registering the raw speckle image does not provide accurate results as the raw speckle image without registering contains disturbances due to the heartbeat and/or respiration of animal. these images are analyzed for contrast value using conventional LASCA (Laser Speckle Contrast Analysis). After obtaining the pre-processed image the next step is to remove the disturbances due to noise from the heartbeat and/or respiration of the animal which can be shown as. Thus rLASCA is performed on the rat’s brain and the results of which are discussed below. After interpolation the alleviated and can be shown as. even small vessels from the pre-processed image. But when using registration of raw speckle image these errors are rectified and it also provides the clear vessel structure where even small vessels can easily be identified. Anna University of Technology. Figure 3: Figure showing the disturbances LASCA is a traditional way of analyzing a raw speckle image. Communications and VLSI Design (NCSCV ’11) 6th & 7th May 2011 Figure 2: Pre-processed Image After obtaining the registered raw speckle images. The raw speckle image and the preprocessed image using convolution kernel are shown as: Figure 1: Raw Speckle Image As we can see a over-exposed spot in the speckle image is disturbed due the heartbeat of the animal which resulted in movement of that spot. Registered image is split into NxN blocks 2. To do that first the rigid translation metric is found using normalized correlation metric and then interpolated using cubic B-spline interpolator. B) PERFORMING CONVENTIONAL LASCA BASED ON REGISTERED IMAGE We can clearly see the vessel structures. The conventional LASCA involves following steps: 1.

Greenberg. G. no. vol. and S. FL: CRC. S. Apr. [11] J. vol. B. Dunn. vol 46.” J. 174–179.. 8. 2004. D.Miao. Jones. Cereb. 2. We implemented a modified registration technique containing a 3×3 convolution kernel. • • Analysis of growth of blood vessels. speckle and related techniques for blood perfusion mapping and imaging. 2004. IV. Bezerianos. Apr. “Imaging the cerebral blood flow with enhanced laser speckle contrast analysis (elasca) by monotonic point transformation. no.” J. CONCLUSION In this paper.. [9] H. Lu. pp. pp. 22. 195– 201. Moskowitz. The next step is to use this registered image to analyze using LASCA. vol. [10] J.. 17. no. Hawkes. Boca Raton.M. Medical Image Registration. H. 1911–1917. Neural Syst. Snyman. “Laser speckle contrast analysis (LASCA): A no scanning. Biomed. Hill. we propose rLASCA to eliminate the effect of disturbances due to respiration and heart beating during imaging the rat’s CBF. full-field technique for monitoring capillary blood flow. [6] D. Y. S. 2003. 151–156. 4..” Appl. Zhu. rLASCA improves the spatial resolution of contrast image and alleviates the noisy effect of disturbances. pp. [8] P. Luo.. Figure 5: Contrast Pattern of single registered image • Analysing brain tumours. Opt.A. and D.. 1996. and S. 3. vol.” Neurosci. C. Biomed. Y. [4] T. REFERENCES [1] J. “Influences of hypothermia on the cortical blood supply by laser speckle imaging’s.Yu. 2009. 2.3rd National Conference on Signal Processing. Burnett. no.. Lett. Boas. 56. P. 21.. Zhou. Furuya. 2005. 128–134. Coimbatore . Blood Flow Metab. Cui. J.Durduran. vol. Q. 368. vol. and. 2009. Eng. Tong. “Modified laser speckle imaging method with improved spatial resolution.. Eng. Anna University of Technology. H. In the application for investigating angiogenesis of rat’s brain tumour. no. 35–66. J.Y. Gong. V. Meas. Biomed. 518–525. Cereb. M. Briers. and H. Hajnal. Blood Flow Metab.” IEEE Trans. [5] M. Practical Mathematical Optimization: An Introduction to Basic Optimization Theory and Classical and New Gradient-Based Algorithms. vol. 10. [2] A. “Heterogeneous response of cerebral blood flow to hypotension demonstrated by laser speckle imaging flowmetry in rats. Cheng. Chen.” Physiol. normalized correlation metric and cubic B-spline interpolator to register the raw speckle images accurately. [3] A. pp. H.Weng. Qiu. After that a false colour map of the above contrast pattern image is taken. “Laser Doppler. First single registered image is taken and for that contrast pattern is found using intensity and standard deviation of that image. Li. 2001. Zhu. Rehabil. Tong.M. 2. The contrast patter of a single registered image without the colour map can be shown as. and J. no. pp.W. Zeng. Opt. “Monitoring thermal-induced changes in tumour blood flow and micro vessels with laser speckle contrast imaging. 2007.” J. 1127–1133. Kharlamov. A. 2001. D. Detre. “Dynamic imaging of cerebral blood flow using laser speckle.” J. Cen. APPLICATION The calculations and the enhanced images obtained using rLASCA are used in various applications such as. 4. J. K. Li. and D. pp. “Spatiotemporal quantification of cerebral blood flow during functional activation in rat somatosensory cortex using laser-speckle flowmetry. New York: Springer-Verlag. Briers and S. Easley. 1. pp. 24. Analysis of cerebral blood flow under various physiological condition of a living organism. Webster.” IEEE Trans. Fontenelle. Qiu. rLASCA shows the small vessels with a higher resolution.Yodh. pp. Brown. Miao. Y. pp. Luo. no. Opt. and S. Yu. Bolay. Communications and VLSI Design (NCSCV ’11) 6th & 7th May 2011 As we can see from the above figure the disturbances are alleviated and the image is thus registered. Figure 6: Contrast pattern of single registered image with colour map The contrast value from this image is calculated and similarly the contrast pattern with colour map is found for the subsequent image and the difference in the contrast value of the two images will give the value of velocity of the blood flow in our region of interest. Organized by Department of ECE. 2001. vol. [7] J. 559–564.

Staal. 739–758.. T. Van Ginneken.” IEEE Trans. Viergever. Imag. vol. Kenyon.M. Communications and VLSI Design (NCSCV ’11) 6th & 7th May 2011 [12] P. Anna University of Technology.M. vol. MI-2. Niemeijer. vol. 7. and B. 2004. Blu. “Comparison of interpolating methods for image resampling. Unser. Imag. Med. pp. Coimbatore . Mar. Parker. Jul.” IEEE Trans. M.. 4. Thevenaz. Troxel. 1983. no. “Interpolation revisited. 31–39. Organized by Department of ECE. Abramoff. Med.. pp.3rd National Conference on Signal Processing. 19. R. [14] J.” IEEE Trans. Med. and D. 23. pp. Imag. no. no. 2000. 1. Apr. 501–509. andM. “Ridge-based vessel segmentation in color images of the retina. [13] J.