Blood Vessel Detection in Images from Laser-Heated Skin Abstract Alireza Kavianpour, Simin Shoari, Behdad Kavianpour CEIS Dept. DeVry University, Pomona, CA 91768 A computer method for recognizing blood vessels in an image constructed by infrared tomography is proposed. Blood vessels detection is important for efficient clinical treatment of a patient. A blood vessel can be model as a block in an image and hence block detection algorithm is applied on an image constructed by infrared tomography. The approach is based on the parallel calculation on hypercube. Hypercube architecture of dimension n is fine-grain architecture with 2 n processors each holding a single pixel of the image. The hypercube operates in an SIMD mode. Two algorithms for computing blocks are explained. Key Words: Block detection, blood vessel, hypercube, image processing, and parallel processing. 1 Introduction Three dimensional reconstructions of the blood vessels in an image taken from part of skin by infrared tomography is an important clinical problem. For example efficient clinical results for laser therapy is based on the selection of proper pulse durations of laser [1,2,3,9] Milner et al. [7] have shown that for efficient treatment of Port of Wine Stain (PWS) (patients with birth's mark), or removing tattoos, laser pulse durations should be approximately equal to the thermal relaxation times of the targeted PWS blood vessels. In this paper we utilize an infrared detector to measure temperature changes induced in a skin exposed to laser radiation. Heat generated due to light absorption by skin diffuses to the surface and results in increased infrared radiation emission levels. By collecting and concentrating the emitted radiation on to an infrared detector, useful information regarding the tested skin will be derived and will be used for blood vessels detection. Vessel diameter varies greatly from patient to patient and even from site to site in one patient. The value of thermal relaxation times, tau_r, is directly proportional to the square of the diameter (d) of the tested vessel and inversely proportional to the thermal diffusivity of skin, tau_r =d^2/16chi where chi is thermal diffusivity of skin (0.9x10-3 cm 2 /s). The Beckman Laser Institute and Medical Clinic in Irvine, California has a laser that user can specify pulse duration over the range of 0.25-15 msec. With such a laser, proper selection of pulse duration of laser exposure for patients is important. In this paper we model blood vessels as blocks and hence block detection algorithm will be used. 2 Experimental Results All the experimental data used in this paper have been accomplished at The Beckman Laser Institute and Medical Clinic. In this center a 0.45 msec pulsed laser source (lambda = 585) with adjustable pulse durations is available. A high-speed Infrared Focal Plane (IR-FPA) camera system takes image of individual laser heated blood vessels. This camera acquires 217 infrared emission frames per second. The infrared signals collected by each detector element are digitized by 3.5 MHZ, 12-bit A/D converter and results are stored in computer. Figure 1 represents infrared tomography instruments. Figure 1 Infrared Radiometery instrumentation
Each frame has 128x128 pixels. Infrared tomography uses a fast infrared focal plane array to detect temperature rises in a substrate induced by pulse radiation. In practice, a pulsed laser is used to produce transient heating of the object under study. The temperature rise, due to the optical absorption of the pulse laser light, creates an increase in infrared emission which is measured by a fast IR-FPA. If a pulse laser source is used to irradiate the skin, an immediate increase in infrared emission will occur due to optical absorption by hemoglobin contained within the blood vessels. An Infrared tomography record of a skin in response to pulsed laser exposure is composed of a sequence of infrared emission frames that indicate localized heating of blood vessels in tested skin. The value of each pixel is presented by P(x,y,t), where x,y are coordinates and t is the measured time sequence. Sample rate is 217 frames per second. 3 Block Detection Algorithm In this section we present an algorithm for detecting connected black pixels. The input to this algorithm is n by n binary pixels. Connectivity among pixels can be defined in terms of their adjacency. Two black pixels (x 1, y 1 ) and (x 2, y 2 ) are 8-neighbor if: Max{ x 1 -x 2, y 1 -y 2 }<=1 and 4-neighbor if: x 1 -x 2 + y 1 -y 2 <= 1 Two black pixels (x 1, y 1 ) and (x k, y k ) are said to connected by 8-path (4-path) if there exists a sequence of black pixels (x p, y p ), 2<= p<=k, such that each pair of pixels (x p-1, y p-1 ) and (x p, y p ) are 8- neighbors (4-neighbors). A maximal connected region of black pixels is called a connected block. We assume a block represents a blood vessel. In this paper we use 8-neighbors method. 3.1 Depth calculation of vessels In this section depth of a vessel z will be calculated. Assume P is a set of pixels indicating vessel V. From 150 frames a curve indicating the relation between temperature change and time will be drawn. The time difference between initial jump and maximum point of this curve is called delayed thermal peak t d.using the equation z=sqrt{4chi t d }, depth of a the vessel V will be calculated, where chi is thermal diffusivity of skin 0.9x10^-3 cm^2/s. 4 Hypercube Architecture Hypercube multi-computer systems have become a subject of considerable interest to the system designers faced with challenging applications. An n-dimensional hypercube multicomputer system, or an n-cub for short, contains 2 n processors each of which is a self-contained computer with its own local memory. Each processor is assigned a unique n-bit address. Two processors are linked if and only if their addresses differ in exactly one bit position. Therefore, each processor has direct communication links to n other processors [4,5,6]. A hypercube computer with the dimension n is Single Instruction Multiple Data ( SIMD) machine. All of the processing elements in the hypercube operate in a strict SIMD mode under the direct control of a single node. Each processing element has its own memory and all of the communication links are bidirectional. The hypercube topology has been proposed as architecture for high-speed image processing where its simple geometry adapts naturally to many types of problems. 5 Simulation Program Application software called Application Visualization System (AVS) is used to process input images. Input images consist of 150 frames. The first few frames and several of last frames will not be used for processing since the early frames are totally black and last frames are white. In order to get better images with less blur and noise, longitudinal inversion algorithm is used. The new gray-level images are transforming into black and white (binary) images. 5.1 Network on AVS: To implement algorithms on binary images a network on AVS with following blocks are defined: File Description: This file reads 150 frames of input images each of size 128x128. Orthogonal Slicer: By setting a parameter one of the 150 images is selected. Crop: This file trims borders of selected image.
Clamp and Contrast: This file changes the graylevel image to black and white or binary images. The format of new image has been changed in order to be readable by processing program. White Image: Stores the image file on the disk. 5.2 Processing Program: The simulation program for detecting blocks was written in C language. Figures 2 and 3 illustrate the result of block detection algorithm on a binary images. (b) Frame 1 (a) Raw image Frame 4 Figure 2. Different frames from 3D tomographic CAM image
Table 2 Results of simulation for block detection algorithm for block B 5 Table 3 Results of simulation for block detection algorithm for block B i Figure 3. Detected blocks in frame 1 Table 1 illustrates the computer simulation results for 128x128 size binary images using Block Detection Algorithm for frame number nine. Table 2 illustrates the computer simulation results for 128x128 size binary images using Block Detection Algorithm for block number B 5. Table 3 illustrates the computer simulation results for 128x128 size binary images using Block Detection Algorithm for Different blocks. Table 1 Results of simulation for block detection algorithm for frame #9 6 Summary In this paper blood vessels are model as blocks. Parallel algorithms for detecting blood vessel on hypercube architecture are described. The result of simulation proves the usefulness of block detection algorithms in image processing and pattern recognition. 7 References [1] Elisa R, Renzo P Retianl Blood Vessel Segmentation Using Line Operators and Support Vector IEEE Trans Medical imaging, 26:1357-1365, 2007 [2] M. Bern, Laser Surgery," Scientific American, pp. 84-90, June 1991
[3] G. Bongiovanni, C. Guerra, and S. Levialdi, Computing the Hough Transform on a Pyramid Architecture," Machine Vision and Applications, vol. 3(2), pp. 117-123, 1990 [4] A. Kavianpour, and N. Bagherzadeh, Circle Detection in Black and White Images, ''Patent pending, UC Case no. 91-195-1, 1991 [5] A. Kavianpour, and N. Bagherzadeh, Finding Circular Shapes in an Image on a Pyramid Architecture, ''Pattern Recognition Letters, vol. 13, no. 12, pp. 843-848, December 1992 [6] A. Kavianpour, and N. Bagherzadeh, Parallel Line Detection for Image Processing on a Pyramid Architecture'' Journal of Pattern Recognition and Artificial Intelligence} vol. 8, no. 1, pp. 37-349, 1994 [8] S. Kime, L. O. Svaasand, M. H. Wilson, M. J. Schell, T. E. Milner, S. Nelson, and M. W. Berns Differential Vascular Response to Laser Photo thrombolysis, "The Journal of Investigative Dermatology vol. 103, no. 5, pp. 693-700, 1994 [9] Cemil Kirbas and Francis Quek A Review of Vessel Extraction Techniques and Algorithms Vision Interfaces and Systems Laboratory (VISLab), Department of Computer Science and Engineering,Wright State University, Dayton, Ohio, January 2003 [10] Bankhead P, Scholfield CN, McGeown JG, Curtis TM Fast Retinal Vessel Detection and Measurement Using Wavelets and Edge Location Refinement. PLoS ONE 7(3): e32435. doi:10.1371/journal.pone.0032435, 2012 [7] T. E. Milner, B. Anvari, S. Nelson, B. S. Tanenbaum, and L. O. Saasand Imaging Laser-Heated Cutaneous Chromospheres," OSA Preceding on Advanced in Optical Imagining and Photon Migration vol. 21 pp. 269-271, 1994