Automatic separation of arteries and veins in optical cerebral cortex images is important in clinical practice and preclinical study. In this paper, a simple but effective automatic artery-vein separation method which utilizes single-wavelength coherent illumination is presented. This method is based on the relative temporal minimum reflectance analysis of laser speckle images. The validation is demonstrated with both theoretic simulations and experimental results applied to the rat cortex. Moreover, this method can be combined with laser speckle contrast analysis so that the artery-vein separation and blood flow imaging can be simultaneously obtained using the same raw laser speckle images data to enable more accurate analysis of changes of cerebral blood flow within different tissue compartments during functional activation, disease dynamic, and neurosurgery, which may broaden the applications of laser speckle imaging in biology and medicine.
© 2011 OSA
Separating arteries and veins in optical images is considerably important to basic researches, clinical diagnostics, and intraoperative procedures: such as studying cerebral blood supply  and angiogenesis , evaluating the changes in vascular structure caused by diabetes and hypertension [3,4], helping surgeons to locate the aneurysm and avoid vascular beds. For practical application, it is necessary to develop a fast, effective and automatic method to separate these two types of vessels instead of introducing empirical knowledge such as the anatomic structure of blood vessels, because empirical knowledge is dependent on individual experience and may lead to serious misjudgment. Akita et al. and Yu et al. took advantage of the different reflect optical intensity between arteries and veins in retinal fundus image to identify the types of vessel [5,6]. Li et al. utilized a piecewise Gaussian model to describe the optical intensity distribution of vessel profile and differentiated retinal arteries and veins by considering the central optical reflecting characteristic of the vessels and estimating the parameters of the piecewise Gaussian model . However, in many cases, the optical intensity profile alone is insufficient to differentiate arteries and veins in single wavelength optical images, since their intensity distributions might be comparable . Therefore multi-wavelength methods were proposed. Schiessl et al. and Vanzetta et al. utilized the ratio of optical density at 540 nm and 560 nm to differentiate arteries and veins [9,10]. Narasimha-Iyer et al. combined structural and functional features to identify the vessel types from dual wavelength images . Luo et al. utilized the t-tests to differentiate arteries and veins based on the changes in deoxygenated hemoglobin (∆[HbR]) computed from dual-wavelength images during transient forebrain ischemia . Zhang et al. utilized photoacoustic microscopy to image variations of hemoglobin oxygen saturation and differentiate the arteries and veins by four optical wavelengths . Besides the multi-wavelength methods, many new methods have been reported in recent years. Hu et al. presented an artery-vein separation method in cerebral cortical images recorded with optical intrinsic signal imaging based on the fact that the spectral distribution of intrinsic physiological oscillations varies from arteries to veins , but this method is complex because it utilizes multi-taper method (MTM), statistical F-test, fuzzy c-means (FCM) and region-growing approach. Song et al. presented an arteriole-venule separation method which was based on the section-illumination photoacoustic microscopy according to the wash-in dynamics of Evans Blue (EB) . Miao et al. proposed an arteries-veins separation method which segmented vascular structure from laser speckle contrast image, and distinguished arteries and veins by anatomical features and optical intensity profiles . However, small artery and vein vessels may be misclassified due to similar absorption to the cortex using this method. MRA and CT techniques have been widely used to differentiate arteries and veins in the head, neck, lungs, heart, and so on [16,17], but these two methods may be unsuitable to be applied to process 2-D optical imaging data because of their specificity related to MRA .
Laser speckle contrast analysis (LSCA) method is a non-scanning full-field optical technique which has been widely used for measuring blood flow changes in brain [18–24], skin , knee , and retina . Speckle contrast is an effective estimator of tissue perfusion, which can be calculated in either spatial or temporal domain [28–30]. With coherent light illumination, it has shown that an enhancement in the cerebral vascular image can be achieved with laser speckle temporal contrast analysis (LSTCA) compared with the vascular images obtained under the illumination of white light or green light . Long-exposure laser speckle contrast analysis imaging enables functional vascular density mapping without the need for exogenous intravascular contrast agents . Kalchenko et al. used dynamic light scattering microscopy which was based on the laser speckle temporal contrast analysis to image tumor blood vessels . However, small arteries and lager veins cannot be differentiated directly from the laser speckle contrast images, because the speckle contrast is related to the blood flow, and there may be similar blood flow velocity in small arteries and large veins.
In this paper, a simple but effective automatic artery-vein separation method which utilizes single-wavelength coherent illumination is presented. This method is based on the relative temporal minimum reflectance (RTMR) analysis of laser speckle images. It utilizes the same experimental setup and data acquisition protocol as laser speckle contrast imaging of blood flow [29,34]. RTMR is defined as the ratio of the temporal minimum intensity to the spatially averaged intensity of the cortical parenchyma neighborhood (for a specific pixel (x, y), the cortical parenchyma neighborhood is a square neighborhood without vessels in the temporal mean speckle image). Our theoretic simulations demonstrate that RTMR is an integrated parameter of reflectance and blood flow velocity. The arteries-veins separation is based on the fact that RTMR value varies among arterial regions, venous regions, and parenchyma. Moreover, a preliminary application of combining this method with LSCA in investigating the blood flow changes in arteries, veins and parenchyma during cortical spreading depression (CSD) is also presented.
The relative temporal minimum reflectance is defined as:
where, RTMR(x, y) is the relative temporal minimum reflectance of time-integrated speckle intensity for a specific pixel (x, y) in a temporal speckle images sequence, Imin,x,y is the temporal minimum intensity of pixel (x, y) defined as the minimum value of the optical intensity at the pixel (x, y) over the temporal laser speckle images sequence recorded by CCD camera, and 〈Icn(x, y)〉s is the spatially averaged intensity of the cortical parenchyma neighborhood of the same pixel in the temporal mean speckle image (the aim of utilizing spatially averaging process is to suppress the experimental noise). The temporal mean speckle image is computed by averaging the sequence speckle images.
In the animal experiments, RTMR image is computed by the temporal minimum intensity image divided by the estimated cortical background image, pixel by pixel. The estimated cortical background image is the resultant image of interpolating each vascular pixel by the spatially averaged intensity of its cortical parenchyma neighborhood in the temporal mean speckle image.
2.1 Temporal minimum intensity
In a typical laser speckle imaging experiment, a laser beam with a wavelength ranging from 600 to 800 nm is used, and a camera exposure time (T) of 5 to 20 ms which is larger than the optical intensity correlation time τ0 is chosen, the minimal time interval ∆t between sequential frames is larger than τ0, so that the speckle signals along the time dimension are uncorrelated and thus are statistically independent [35–37]. Although, the probability density function (PDF) of the intensity recorded by the CCD is approximated to a Gamma function , some researchers have presented that the intensity variations display a Rayleigh distribution instead of an exponential distribution under laser speckle with unpolarized illumination in many biological applications and coherent optical imaging modalities [39–41].
In this paper, for the convenience of mathematical derivation, the Rayleigh distribution function is utilized as an approximation of the Gamma distribution function for the temporal intensity variation of the integrated dynamic speckle under certain specific conditions: the speckle contrast value is very small, and the time sequential speckle images are statistically independent. The theoretical analysis, numerical simulation and phantom experimental data are provided to validate this approximation as following.
Theoretical analysis: the formula of the Gamma function can be expressed as :
and the formula of the Rayleigh function can be expressed as :
where, p and pg are the intensity PDF, I is the intensity, μ is the mean intensity, and σ is the standard deviation of the same data set. The parameters a g, b g, a and b are four symbolic constants. And, b g is the number of degrees of freedom within the integrated laser speckle . When b g is very small, the Gamma distribution approaches a negative exponential distribution, while the Gamma distribution approaches a Gaussian distribution when b g is big enough (infinity) .
For convenience, we set μ = 1, and compare the Gamma distribution and the Rayleigh distribution under different σ. The normalized results are shown in Fig. 1(a) : when σ/μ is big (σ/μ = 0.9, 0.7), there exist significant differences between Gamma distribution and Rayleigh distribution; When σ/μ is small (σ/μ = 0.5, 0.3 and 0.1), the PDF of Rayleigh distribution approaches to that of Gamma distribution.
Numerical simulation: two data sets of simulated dynamic laser speckle images with two different laser speckle contrasts were obtained by numerical simulation [43,44], and each data set had 500 frames image (100 × 100). For each data set, a random selected pixel was obtained (named as A), the temporal PDF of A was computed along the 500 frames and plotted in Fig. 1(b) by solid line. The fitted Rayleigh distribution curve of A was obtained from the same data set by the least square method, and was plotted in Fig. 1(b) by dashed line.
The moving white plate phantom experiments: a white plate moved by a step motor was placed in the laser speckle imaging system (as the same system described in Methods and Material section). Two data sets (each had 500 frames image) with two different speeds were recorded by CCD camera (with 20 ms exposure time). For each data set, the same data processing methods in the latest paragraph were used and the results were plotted in Fig. 1 (c).
From the previous literature, most of the speckle temporal contrast Kt(x, y) values of dynamic laser speckle images are very small in biomedical applications of laser speckle contrast analysis [18,21,29,30,45]. Based on these results, we can utilize the Rayleigh distribution function as an approximation of the Gamma distribution function for the applications of biological tissue.
In Eq. (3), we redefine It,x,y as the intensity at pixel (x, y) in the tth frame, μx,y is the mean intensity computed from the intensities at the same position (x, y) along the sequence laser speckle images, and σx,y is the standard deviation of the intensities at the same position along the same data set. Here, we only consider the condition of It,x,y ≥ a. The following equation can be obtained from Eq. (3),
We define two symbolic variables w and X as shown in Eq. (5):
Obviously, w decreases with the increase of (It,x,y – a)2. And Eq. (4) can be simplified as:
Equation (6) is a Lambert’s W function. The PDF of the temporal minimum intensity at pixel (x, y) must be very close to zero because the temporal intensity probability density function obeys Rayleigh distribution. And b is a positive value constant. Therefore, X is a negative real value and very close to zero. In this case, there are two possible negative real value solutions of w for the Lambert’s W function, one is approximate to zero and another is much smaller than −1 [46,47]. Because w decreases with the increase of (It,x,y – a)2 and It,x,y ≥ a, the solution of w for the temporal minimum intensity is the bigger one which is approximate to zero. According to Eq. (3) and Eq. (5), the temporal minimum intensity (Imin,x,y) at pixel (x, y) in the sequence laser speckle images can be estimated by the following equation:
Laser speckle temporal contrast is defined as :
where, <Ix,y>t is the average intensity at the same position along the sequence laser speckle images (<Ix,y>t is the same to μx,y), and Kt(x, y) is the laser speckle temporal contrast of the same data set.
According to Eq. (9), it can be seen that the temporal minimum intensity is an integrated variable of the temporal averaged intensity of reflect light and temporal laser speckle contrast. Therefore, the temporal minimum intensity can show a mixed effect on the absorption of chromophore in tissue and blood flow velocity.
The validation of Eq. (9) is demonstrated through comparing the theoretical predicted temporal minimum intensity image obtained by employing Eq. (9) with that calculated directly from the experimental data. Figure 2(a) shows the temporal minimum intensity image computed from 500 sequence laser speckle images. Figure 2(b) shows the theoretical predicted temporal minimum intensity image, where the average intensity <Ix,y>t and the laser speckle temporal contrast Kt(x, y) at each pixel are computed from the same data set as for Fig. 2(a). Figure 2(c) shows the relative difference image between Fig. 2(a) and Fig. 2(b), which is defined as |(Fig. 2(b) - Fig. 2(a))/Fig. 2(a)|, and the maximum relative error between the experimental result and the theoretical predicted one is smaller than 2%. Moreover, as shown in Fig. 2(a) and Fig. 2(b), vessels can be classified into two groups, one group with higher Imin,x,y than other cortical compartments and another group with lower Imin,x,y than it’s cortical parenchyma neighborhood. But, there is inhomogeneous background due to the uneven illumination, which decreases the accuracy of classification of two groups. In the following sections, we utilize a parameter named RTMR to eliminate such influence.
2.2 Relative temporal minimum reflectance (RTMR)
To avoid the influence from inhomogeneous background due to the uneven illumination, the relative temporal minimum reflectance (RTMR) is utilized. Substituting Eq. (9) into Eq. (1), the RTMR can be expressed as:
Assuming that the intensity of the incident light is I0, the reflectance at pixel (x, y) is R(x, y) and the average reflectance of the cortical parenchyma neighborhood of pixel (x, y) is Rcn(x, y), then <Ix, y>t/ <Icn(x, y)>s = R(x, y)/Rcn(x, y). Thus, the Eq. (10) can be simplified as:
Equation (11) suggests that RTMR is an integrated parameter of tissue reflectance coefficient and blood flow velocity. Furthermore, RTMR is independent of the intensity distribution of the illumination light, thus the RTMR image will display a much more homogeneous cortical background. In the following sections, we demonstrate the validation of artery-vein differentiation by RTMR analysis.
2.3 Differentiating the arteries and veins by RTMR analysis
Cheng et al. have proposed that 1/Kt2 can be approximately regarded as a proportional to the mean blood velocity [27,35]. Therefore, we constructed a two-dimensional vector (1/Kt2, R/Rcn) to demonstrate the validation of artery-vein differentiation by RTMR analysis.
First, the distribution image of RTMR as a function of R/Rcn and 1/Kt2 was computed with Eq. (11) and drawn in Fig. 3(a) . The pseudo color is used to quantify the RTMR value, blue to red: low RTMR to high RTMR. And, the contour lines map of RTMR was obtained from Fig. 3(a).
Second, an artery-vein identification image was obtained by the ratio of optical density at 540 nm and 560 nm as shown in Fig. 3(b), white to black: arteries, parenchyma, veins [9,10]. Ground-truth data were obtained along the vascular structure in Fig. 3(b) by a physiological expert, who manually picked significant vascular points and indicated their types, and a total of 400 vascular points were identified and marked. 200 arterial points are marked by red color and 200 venous points are marked by blue color.
Third, 1/Kt2 for each of the 400 vascular points was obtained from 500 sequence laser speckle images under He-Ne illumination. R/Rcn values of arteries and veins are related to the wavelengths. Here, several common wavelengths were utilized in animal experiments to find the effective wavelength range for the RTMR analysis. R/Rcn of each point under different wavelength was computed by the reflecting intensity divided by the spatially averaged reflecting intensity of its cortical parenchyma neighborhood without vessel. The vector (1/Kt2, R/Rcn) of each point was plotted in the contour lines maps of RTMR as shown in Fig. 3(c) –Fig. 3(g), arterial vectors are marked by red color and venous vectors are marked by blue color.
As shown in the green shadows in Fig. 3(c)–Fig. 3(g), it’s difficult to differentiate arteries and veins only by the 1/Kt2 because that many points in arteries and in relatively larger veins have similar flow velocity, although some points in relatively larger arteries can be segmented due to much higher velocities. As shown in the grey shadows in Fig. 3(d), Fig. 3(f) and Fig. 3(g), many relatively larger arterial points and many venous points may be misclassified only by the relative reflectance because that these points have similar reflectance, although some points in relatively larger veins can be segmented due to much lower reflectance (much higher absorption). These are the limitations and advantages of artery-vein separation by using the characteristics of reflectance of single wavelength or laser speckle contrast analysis alone.
RTMR analysis merges the advantages of reflectance of single wavelength and laser speckle contrast analysis in artery-vein separation. As shown in Fig. 3(f) and Fig. 3(g), according to the contour lines of RTMR, arterial points and venous points can be differentiated easily. (1): the points in relatively smaller veins and relatively larger arteries can be accurately classified. This is because blood velocities in small veins are much slower than those in relatively larger arteries, although the relative reflectance of those vessels may be similar. (2): the points in arteries and relatively larger veins can be accurately classified. This is because the relative reflectance in relatively larger veins is much lower than those in arteries, although the blood velocities may be similar. (3): the arteries and veins with similar diameters can be differentiated accurately. This is because flow velocities in these arteries are higher than in those veins and the relative reflectance values of these arteries are higher than those veins. Therefore, the RTMR analysis is an effective artery-vein differentiation method.
As has been emphasized, RTMR is related to the relative reflectance and this method is wavelength dependent. In this section, we estimate the effective wavelength range which can be used in RTMR analysis. As shown in Fig. 3(c)–Fig. 3(e), arterial points and venous points cannot be accurately differentiated by RTMR due to high absorption of hemoglobin. Comparing the distributions of the arterial points with those of the venous points in Fig. 3(f) and Fig. 3(g), the deviation of difference between arterial point-envelope and venous point-envelope in Fig. 3(g) is much smaller than that in Fig. 3(f), which is arising from lower absorption of hemoglobin. Therefore, those wavelengths with very high or very low absorption of hemoglobin are not suitable in RTMR analysis. According to the spectrum of hemoglobin , the effective wavelength which can be used in our method is from approximate 600 nm to 640 nm and the optimal wavelength may be 600nm.
According to the analysis in the upper sections, we can estimate the distributions of RTMR of arteries, veins and cortical parenchyma under the effective wavelengths. Figure 3(a) is redrawn in Fig. 3(h). R/Rcn of cortical parenchyma is 1, because reflectance values of cortical parenchyma are the same as those of their neighborhoods without vessels. The blood velocity in cortical parenchyma is very low. Therefore, the RTMR distribution of cortical parenchyma is in the left upper corner as shown in Fig. 3(h), marked with a gray closed curve. According to Fig. 3(f) and Fig. 3(g), the distribution of RTMR of arteries is marked with a white closed curve and the distribution of RTMR of veins is marked with a black closed curve as shown in Fig. 3(h). According to Fig. 3(h), the RTMR values in arteries are higher than other parts, the RTMR values in relatively larger veins are lower than other parts and the RTMR values in relatively smaller veins are similar to those in cortical parenchyma. Arterial regions can be segmented from other parts in the cerebral cortex based on the fact that the RTMR values in arterial regions are higher than those of their cortical parenchyma neighborhoods. To avoid misclassification of relatively smaller veins, the venous regions can be obtained by removing the arterial regions from the vascular structures which are segmented from laser speckle temporal contrast image.
3. Methods and Material
3.1 Imaging system
A beam of He-Ne laser (Melles Griot, America; 632.8 nm, 15 mW) was expanded and collimated to illuminate the sample at about 30° incidence. 500 frames of laser speckle images were acquired continuously by a 12-bit CCD camera (PixelFly QE, PCO Computer, Germany) attached to a stereomicroscope (Olympus SZ6045TR Zoom, Japan) for data processing. The CCD exposure duration was set to 20ms. Light intensity was controlled by a variable attenuator to be within the dynamic range of the CCD camera. The adult male Wistar rats were anesthetized with an intraperitoneal injection of a mixture of α-chloralose (180 mg/kg) with urethane (900 mg/kg) and were fixed in a stereotaxic frame. A craniotomy was made over the left parietal cortex to form an approximately 5.0 mm × 5.0 mm cranial window. All experimental procedures were approved by the Committee for the Care and Use of Laboratory Animals at Huazhong University of Science and Technology.
3.2 Data processing
Before data processing, we utilized numerical simulation of time-integrated dynamic speckle images [43,44] to estimate the optimum parameters for this artery-vein separation method. The simulation results showed that the optimum number of image frames involved in data processing should be larger than 150 to ensure accurate statistical results. To further improve the signal to noise ratio (SNR), it is recommended that the total frames are divided into several groups equally and each group possesses laser speckle images no less than 30 frames. In this study, 500 laser speckle images were employed and divided into 10 groups equally for further data processing.
For simplicity, the vascular map and the estimated cortical parenchyma background image were computed before executing the artery-vein separation. The vascular map was segmented from the laser speckle temporal contrast image (the laser speckle temporal contrast image was computed from 500 laser speckle images) due to their lower speckle contrast values (higher blood flow velocity) in these vessels than in cortical parenchyma. The estimated cortical parenchyma background image was the resultant image of interpolating each vascular pixel by the spatially averaged intensity of its cortical parenchyma neighborhood in the temporal mean speckle image, as shown in Fig. 4(c) . Here, the neighborhood region was a square region, and the width of which was defined slightly bigger (5 – 10 pixels) than the diameter of the biggest vessel due to inhomogeneous background of temporal mean speckle image. The diameter of the biggest vessel was estimated in the vascular map.
Step 1: Computing the temporal minimum intensity image
500 frames of laser speckle image were equally divided into 10 groups, and in each group the temporal minimum intensity image was obtained by estimating the temporal minimum intensity for each pixel along 50 image frames. Then the 10 frames of temporal minimum intensity images were averaged.
Step 2: Computing the relative temporal minimum reflectance image
According to the Eq. (1), the relative temporal minimum reflectance image was computed by the temporal minimum intensity image divided by the estimated cortical background image, pixel by pixel.
Step 3: Differentiating arteries and veins
In the RTMR image, each pixel of vessels was compared with the averaged RTMR value of its cortical parenchyma neighborhood. First, arterial regions were segmented based on the fact that RTMR value of each arterial pixel was higher than its cortical parenchyma neighborhood. Then, the venous regions were obtained by removing the arterial regions from the vascular map. Here, we did not directly segment the venous regions because that the RTMR values in small veins were similar to those in neighboring cortical parenchyma, which could make it difficult to obtained those small veins. Hence, we chose the subtraction strategy to obtain the veins.
3.3 Validation of relative temporal minimum reflectance analysis method
Validation of the relative temporal minimum reflectance analysis was performed through comparing it with the multi-wavelength method. A halogen light (Olympus LG-PS2, Japan) was employed to illuminate the sample. To record the reflected images under different wavelengths, a LCTF (VariSpec VSI, Cri, USA) was placed between the CCD camera and the stereomicroscope. In the experiment, 540 nm, 560 nm, 570 nm, 600 nm, and 632.8 nm wavelengths were selected. CCD camera recorded 500 image frames of the rat cortex at each wavelength. Two images of a white plastic reflectance standard sample (Olympus, Japan), calibrated to be Rstd540 = 0.890 and Rstd560 = 0.891 at 540 nm and at 560 nm in separate experiments, were acquired, and were referred as Istd540 and Istd560, respectively. The optical density and the ratio of optical density at 540 nm and 560 nm were computed as
where, I540 is the mean image of 500 animal images at 540 nm. Istd540 is the reflect image of the white plastic at 540 nm. OD540 is the optical density at 540 nm. I560 is the mean image of the 500 animal images at 560 nm. Istd560 is the reflect image of the white plastic at 560 nm. OD560 is the optical density at 560 nm. ODR is the ratio of the optical density at 540 nm and 560 nm.
4.1 Relative temporal minimum reflectance image
Figure 4(e) is the RTMR image, and there is much more homogeneous background than the temporal minimum intensity image as shown in Fig. 4(b). Figure 4(d) is the artery-vein identification image obtained by the ratio of optical density at 540 nm and 560 nm, white to black: arteries, parenchyma and veins. Comparing Fig. 4(d) with Fig. 4(e), similar structures can be found. The vessels with higher values than the values of their cortical parenchyma neighborhoods in Fig. 4(e) are consistent with the arteries in Fig. 4(d); the vessels with lower values than the values of their cortical parenchyma neighborhoods in Fig. 4(e) are similar to the veins in Fig. 4(d), but small veins cannot be found in Fig. 4(e). These animal experimental results are consistent with our prediction.
4.2 Artery-vein separation
The result of artery-vein separation for the whole image by RTMR analysis of single-wavelength laser speckle images is shown in Fig. 4(f). To facilitate comparison, four vascular sections are identified and marked as A1, A2, A3 and A4 as shown in Fig. 4(a). A1 and A2 can be identified easily as arteries from their anatomic structure, and our judgment is demonstrated to be correct as shown in Fig. 4(f). Interestingly, vessel types of the vessels marked as A3 and A4 in Fig. 4(a) may be identified as veins when only empirical knowledge is employed, but Fig. 4(f) shows that both of them are arteries and our judgment is demonstrated to be true as compared with Fig. 4(d).
4.3 Truth Positive Rate (TPR)
To avoid the influence from bones in the margins, a central region is selected and marked by green rectangle as shown in Fig. 4(a). The artery-vein identification image obtained by the ratio of optical density image at 540 nm and 560 nm in the green rectangle has been drawn in Fig. 3(b) and redrawn in Fig. 5(c) . Ground-truth data are the same data set as has been mentioned in the theory section. 200 arterial points and 200 venous points are marked with different color as shown in Fig. 5(c). RTMR value of each point and the averaged RTMR value of its cortical parenchyma neighborhood are computed. The RTMR values of venous points and their cortical parenchyma neighborhoods are plotted in Fig. 5(a), point by point. The RTMR values of arterial points and their cortical parenchyma neighborhoods are plotted in Fig. 5(b), point by point. The number of arterial points which are segmented by our method is 207, and 197 classified points are correct. The number of venous points which are segmented by our method is 193, and 190 classified points are correct. The true positive rate (TPR) is computed by the total number of the correctly classified arterial/venous points divided by the number of ground true arterial/venous points . The sum of the true positive rate and misclassification rate is 1 for one vessel type. The TPR of our separation method reaches 98.5% for the arteries, and 95% for the veins. The misclassification of arteries as veins is 1.5%, and the misclassification of veins as arteries is 5%. For comparison, the TPR of dual-wavelength method (the ratio of optical density at 540 nm and 560 nm) was 96.5% and 98% for the arteries and veins respectively obtained using our experimental data. Narasimha-Iyer et al. utilized dual-wavelength (570 nm and 600 nm) imaging which provided both structural and functional features to differentiate arteries and veins, and the TPR of dual-wavelength imaging was reported to be 97% and 90% for the retinal arteries and veins respectively .
5.1 Wavelength dependency and penetration depth
Our theoretic simulation and animal experimental results suggest that effective wavelength of RTMR analysis is between 600 nm and 640 nm. As shown in Fig. 6 , arteries and veins cannot be differentiated under 540 nm or 660 nm (shown in Fig. 6(b) and Fig. 6(d)). This is because that absorption of hemoglobin is significantly high under 540 nm and significantly low under 660 nm. Figure 6(c) is the temporal minimum intensity image under He-Ne laser illumination, and the temporal minimum intensities vary among arteries, veins and cortical parenchyma. Therefore, arteries can be segmented from Fig. 6(c) and veins can be obtained by removing the arterial structure from the vascular map.
The RTMR analysis method utilizes the same experimental setup and data acquisition protocol as laser speckle contrast imaging of blood flow. The wavelength of the laser light, the tissue optical properties, and the source-detector geometry determine the penetration depth of laser speckle contrast imaging. Dunn et al. indicated that the penetration depth of speckle images is comparable to that of laser-Doppler flowmetry . Jakobsson and Nilsson reported that the penetration depth of laser-Doppler flowmetry was between 500 μm and 1 mm . According to the Monte Carlo simulation on the brain tissue, the penetration depth of RTMR analysis method is a few hundred microns in brain tissue under the effective wavelengths. Due to that CCD may collects the light from the tissue which is in deeper than the penetration depth of the RTMR analysis method, the vessels which are in deeper than the penetration depth will affect the measured RTMR value. For example, if there has an artery below the penetration depth and the CCD can collect the light from this artery, the lower absorption and higher velocity of this artery will increase the measured RTMR value. Otherwise, if there has a vein, the higher absorption and lower velocity of the vein will decrease the measured RTMR value.
5.2 Comparing with reflectance of single wavelength and laser speckle contrast
Figure 7(a) –(e) are the normalized reflected images under 540 nm, 560 nm, 570 nm, 600 nm and 632.8 nm wavelengths in the same cerebral area as in Fig. 4(a). The relatively larger veins are notable due to the higher absorption in these regions than in the arterial regions and the cortical parenchyma regions. However, this performance has several drawbacks in differentiating arteries and veins. The first drawback: vessel types in the margins of the image cannot be correctly distinguished due to inhomogeneous illumination. The second drawback: as shown in Fig. 7(b), Fig. 7(d) and Fig. 7(e), relatively smaller veins may be misclassified due to similar absorptions in these areas with those in relatively larger arteries because of the lower absorption in arteries with higher oxygen saturation at these wavelengths; as shown in Fig. 7(a), arteries may not be correctly distinguished due to similar absorptions in these areas with those in relatively larger veins because of the higher absorption at this wavelength in arteries with higher oxygen saturation. The speckle temporal contrast is redrawn in Fig. 7(f), and arteries and veins cannot be differentiated directly from the temporal contrast image because arteries and veins with the same blood flow velocities correspond to the same speckle contrast values. These results are consistent with the conclusions in the theory section.
Murari et al. proposed that approximately 20 μm resolution could be obtained when utilizing laser speckle temporal contrast analysis to image vascular structures within a few millimeters across field . Bui et al. proposed that long-exposure (e.g., >1000 ms) laser speckle contrast analysis enabled visualization of blood flow in capillaries and small arterioles and venules . These suggest that a potential possibility of detecting the blood flow changes in arterioles, venules and parenchyma respectively by combining relative temporal minimum reflectance analysis with laser speckle contrast analysis.
Here is an instance of detecting the blood flow changes in different cortical compartments respectively by the combined method during cortical spreading depression (CSD) induced by KCl simulation in a burr hole in the ipsilateral frontal bone. The ratio change of blood flow during CSD, calculated by (V – V0)/V0 (V0 is the flow velocity before CSD, V is the flow velocity during our experiment) is shown in Fig. 8(a) . Traditionally, the types of cortical compartments are identified by the anatomic structure of blood vessels. But such identification is unsuitable to apply to analyze small vessels or the cortical regions with complicated vascular structures. As shown in Fig. 8(b), it’s difficult to classify types of cortical compartments in P because that the vessels are very small (< 20 μm). In the traditional way, the changes of blood flow velocity in P are averaged. The results are shown in Fig. 8(d) marked by green curve. This averaging process reduces the signal accuracy due to that the different changes of blood flow velocity among different cortical compartments in this area are ignored.
When combining laser speckle contrast analysis with relative temporal minimum reflectance analysis, blood flow changes in different types of cortical compartments can be detected respectively. First, arterioles and venules in P are separated from 500 laser speckle images recorded by CCD before CSD, as shown in Fig. 8(c). Then, the changes in the same type of tissue compartment are averaged respectively. This averaging process increases the accuracy of signal and SNR. The ratio changes of blood flow velocity in different types of cortical compartments in P are shown in Fig. 8(d). The peak value of the ratio change of blood flow velocity in cortical parenchyma computed by the combined method is much higher than the traditional method. Here we ignore the change of vascular diameter during the experiment.
The experimental results demonstrate that relative temporal minimum reflectance analysis is a simple and effective method of separating arteries and veins in cerebral cortex, and it can be combined with laser speckle contrast analysis so that the artery-vein separation and blood flow imaging can be obtained simultaneously during ischemia, CSD, and other disease models, although the absorption and scattering of underlying parenchyma and vessels will always influence the accuracy of arteries-veins separation.
In summary, we present a simple but effective automatic artery-vein separation method which utilizes single-wavelength coherent illumination in the cerebral cortex. This method is based on the relative temporal minimum reflectance analysis of laser speckle images. By combining this method with laser speckle contrast analysis method, artery-vein separation and blood flow imaging can be obtained simultaneously using the same raw laser speckle images data to improve the compartment-resolved imaging of cerebral blood flow during functional activation, disease dynamic, and neurosurgery, which may broaden the applications of laser speckle imaging in biology and medicine.
We thank Xuejun Tang, Bing Li, Hongyan Zhang and Zhen Wang for their work in manuscript revision. This work is supported by the Program for New Century Excellent Talents in University (Grant No. NCET-08-0213), the National High Technology Research and Development Program of China (Grant No. 2007AA02Z303), the National Natural Science Foundation of China (Grant Nos. 30970964, 30800339, 30801482, 30800313) and the Ph.D. Programs Foundation of Ministry of Education of China (Grant No. 20090142110054).
References and Links
3. N. E. Cameron and M. A. Cotter, “The relationship of vascular changes to metabolic factors in diabetes mellitus and their role in the development of peripheral nerve complications,” Diabetes Metab. Rev. 10(3), 189–224 (1994). [CrossRef] [PubMed]
4. F. Hansen-Smith, A. S. Greene, A. W. J. Cowley Jr, and J. H. Lombard, “Structural changes during microvascular rarefaction in chronic hypertension,” Hypertension 15(6 Pt 2), 922–928 (1990). [PubMed]
5. K. Akita and H. Kuga, “A computer method of understanding ocular fundus images,” Pattern Recognit. 15(6), 431–443 (1982). [CrossRef]
6. J. J. Yu, B. Hung, and H. Sun, “Automatic recognition of retinopathy from retinal images,” in Proceedings of IEEE Conference on Engineering Medicine and Biology Society (Institute of Electrical and Electronics Engineers, Philadelphia, 1990), pp. 171–173.
7. H. Li, W. Hsu, M. L. Lee, and H. Wang, “A piecewise Gaussian Model for profiling and differentiating retinal vessels,” in Proceedings of IEEE Conference on Image Processing (Institute of Electrical and Electronics Engineers, Barcelona, 2003), pp. 1069–1072.
8. H. Narasimha-Iyer, J. M. Beach, B. Khoobehi, and B. Roysam, “Automatic identification of retinal arteries and veins from dual-wavelength images using structural and functional features,” IEEE Trans. Biomed. Eng. 54(8), 1427–1435 (2007). [CrossRef] [PubMed]
10. I. Vanzetta, R. Hildesheim, and A. Grinvald, “Compartment-resolved imaging of activity-dependent dynamics of cortical blood volume and oximetry,” J. Neurosci. 25(9), 2233–2244 (2005). [CrossRef] [PubMed]
11. Z. Luo, Z. Yuan, Y. Pan, and C. Du, “Simultaneous imaging of cortical hemodynamics and blood oxygenation change during cerebral ischemia using dual-wavelength laser speckle contrast imaging,” Opt. Lett. 34(9), 1480–1482 (2009). [CrossRef] [PubMed]
12. H. F. Zhang, K. Maslov, M. Sivaramakrishnan, G. Stoica, and L. V. Wang, “Imaging of hemoglobin oxygen saturation variations in single vessels in vivo using photoacoustic microscopy,” Appl. Phys. Lett. 90(5), 053901 (2007). [CrossRef]
13. D. Hu, Y. Wang, Y. Liu, M. Li, and F. Liu, “Separation of arteries and veins in the cerebral cortex using physiological oscillations by optical imaging of intrinsic signal,” J. Biomed. Opt. 15(3), 036025 (2010). [CrossRef] [PubMed]
15. P. Miao, M. Li, N. Li, A. Rege, Y. Zhu, N. Thakor, and S. Tong, “Detecting cerebral arteries and veins: from large to small,” J. Innovative Opt. Health Sci. 03(01), 61–67 (2010). [CrossRef]
17. J. Svensson, P. Leander, J. H. Maki, F. Stahlberg, and L. E. Olsson, “Separation of arteries and veins using flow-induced phase effects in contrast-enhanced MRA of the lower extremities,” Magn. Reson. Imaging 20(1), 49–57 (2002). [CrossRef] [PubMed]
19. A. K. Dunn, A. Devor, A. M. Dale, and D. A. Boas, “Spatial extent of oxygen metabolism and hemodynamic changes during functional activation of the rat somatosensory cortex,” Neuroimage 27(2), 279–290 (2005). [CrossRef] [PubMed]
20. J. S. Paul, A. R. Luft, E. Yew, and F. S. Sheu, “Imaging the development of an ischemic core following photochemically induced cortical infarction in rats using Laser Speckle Contrast Analysis (LASCA),” Neuroimage 29(1), 38–45 (2006). [CrossRef] [PubMed]
21. T. P. Obrenovitch, S. Chen, and E. Farkas, “Simultaneous, live imaging of cortical spreading depression and associated cerebral blood flow changes, by combining voltage-sensitive dye and laser speckle contrast methods,” Neuroimage 45(1), 68–74 (2009). [CrossRef] [PubMed]
22. E. Farkas, F. Bari, and T. P. Obrenovitch, “Multi-modal imaging of anoxic depolarization and hemodynamic changes induced by cardiac arrest in the rat cerebral cortex,” Neuroimage 51(2), 734–742 (2010). [CrossRef] [PubMed]
23. Z. Wang, W. Luo, P. Li, J. Qiu, and Q. Luo, “Acute hyperglycemia compromises cerebral blood flow following cortical spreading depression in rats monitored by laser speckle imaging,” J. Biomed. Opt. 13(6), 064023 (2008). [CrossRef] [PubMed]
24. Z. Luo, Z. Yuan, M. Tully, Y. Pan, and C. Du, “Quantification of cocaine-induced cortical blood flow changes using laser speckle contrast imaging and Doppler optical coherence tomography,” Appl. Opt. 48(10), D247–D255 (2009). [CrossRef] [PubMed]
25. B. Choi, N. M. Kang, and J. S. Nelson, “Laser speckle imaging for monitoring blood flow dynamics in the in vivo rodent dorsal skin fold model,” Microvasc. Res. 68(2), 143–146 (2004). [CrossRef] [PubMed]
26. R. C. Bray, K. R. Forrester, J. Reed, C. Leonard, and J. Tulip, “Endoscopic laser speckle imaging of tissue blood flow: applications in the human knee,” J. Orthop. Res. 24(8), 1650–1659 (2006). [CrossRef] [PubMed]
28. J. D. Briers and S. Webster, “Laser speckle contrast analysis (LASCA): a nonscanning, full-field technique for monitoring capillary blood flow,” J. Biomed. Opt. 1(2), 174–179 (1996). [CrossRef]
30. P. Li, S. Ni, L. Zhang, S. Zeng, and Q. Luo, “Imaging cerebral blood flow through the intact rat skull with temporal laser speckle imaging,” Opt. Lett. 31(12), 1824–1826 (2006). [CrossRef] [PubMed]
32. A. K. Bui, K. M. Teves, E. Indrawan, W. Jia, and B. Choi, “Longitudinal, multimodal functional imaging of microvascular response to photothermal therapy,” Opt. Lett. 35(19), 3216–3218 (2010). [CrossRef] [PubMed]
33. V. Kalchenko, D. Preise, M. Bayewitch, I. Fine, K. Burd, and A. Harmelin, “In vivo dynamic light scattering microscopy of tumour blood vessels,” J. Microsc. 228(2), 118–122 (2007). [CrossRef] [PubMed]
34. W. Luo, P. Li, Z. Wang, S. Zeng, and Q. Luo, “Tracing collateral circulation after ischemia in rat cortex by laser speckle imaging,” J. Innovative Opt. Health Sci. 01(02), 217–226 (2008). [CrossRef]
35. H. Cheng, Y. Yan, and T. Q. Duong, “Temporal statistical analysis of laser speckle images and its application to retinal blood-flow imaging,” Opt. Express 16(14), 10214–10219 (2008). [CrossRef] [PubMed]
36. P. Zakharov, A. C. Völker, M. T. Wyss, F. Haiss, N. Calcinaghi, C. Zunzunegui, A. Buck, F. Scheffold, and B. Weber, “Dynamic laser speckle imaging of cerebral blood flow,” Opt. Express 17(16), 13904–13917 (2009). [CrossRef] [PubMed]
37. J. W. Goodman, Statistical Optics (Wiley & Sons, New York, 1985).
38. J. W. Goodman, Speckle Phenomena in Optical: Theory and Applications (Roberts and Company, Englewood, Colorado, 2007).
39. D. D. Duncan and S. J. Kirkpatrick, “Performance analysis of a maximum-likelihood speckle motion estimator,” Opt. Express 10(18), 927–941 (2002). [PubMed]
40. S. J. Kirkpatrick, D. D. Duncan, R. K. Wang, and M. T. Hinds, “Quantitative temporal speckle contrast imaging for tissue mechanics,” J. Opt. Soc. Am. A 24(12), 3728–3734 (2007). [CrossRef] [PubMed]
42. R. C. Gonzalez, R. E. Woods, and S. L. Eddins, Digital Image Processing Using MATLAB (Prentice Hall, New York, 2004).
44. J. Qiu, P. Li, W. Luo, J. Wang, H. Zhang, and Q. Luo, “Spatiotemporal laser speckle contrast analysis for blood flow imaging with maximized speckle contrast,” J. Biomed. Opt. 15(1), 016003 (2010). [CrossRef] [PubMed]
45. J. C. Ramirez-San-Juan, R. Ramos-García, I. Guizar-Iturbide, G. Martínez-Niconoff, and B. Choi, “Impact of velocity distribution assumption on simplified laser speckle imaging equation,” Opt. Express 16(5), 3197–3203 (2008). [CrossRef] [PubMed]
46. R. M. Corless, G. H. Gonnet, D. E. G. Hare, and D. J. Jeffrey, “Lambert's W function in Maple,” Maple Tech. Newslett. 9, 12–22 (1993).
47. R. M. Corless, G. H. Gonnet, D. E. G. Hare, D. J. Jeffrey, and D. E. Knuth, “On the Lambert-W Function,” Adv. Comput. Math. 5(1), 329–359 (1996). [CrossRef]
48. S. Prahl, “Optical Absorption of Hemoglobin,” (1999), http://omlc.ogi.edu/spectra/hemoglobin/index.html.