2.1.

Use of Human Subjects

The occlusion protocol and neonatal imaging measurements were approved by the Institutional Review Board of University of California, Irvine (ClinicalTrials.gov identifier NCT01484730 and NCT01483703, respectively).

2.2.

Handheld Laser Speckle Imaging Device

Our LSI device (Fig. 1) consisted of a monochrome camera, laser pointer, and tablet computer. The monochrome camera was an 8-bit, 1.32 megapixel ($3.75\text{-}\mu \mathrm{m}\times 3.75\text{-}\mu \mathrm{m}$ pixel size) CCD camera (CMLN-13S2M-CS, Point Grey, Richmond, Canada) equipped with a close-focus zoom lens (MLH-10x, Computar, CBC America Corp, New York). We attached the camera to the back of a Surface Pro 2 tablet (Microsoft, Redmond, Washington). For in vitro experiments, we attached a 650-nm center wavelength bandpass filter (Edmund Optics, Barrington, New Jersey) to the lens of the camera to mitigate the effects of room lighting on the collected images. For in vivo experiments, we performed the measurements with all room lights turned off. The pixel resolution of the images was $\sim 120\text{pixels}/{\mathrm{mm}}^2$.

Fig. 1

(a) Tablet-based LSI system. (b) Representative example of system in use to collect LSI images from a human subject.

For the excitation source, we used two lasers. For in vitro and clinical experiments, we used a 650-nm laser pointer (5 mW, Five Star Inc., West Chester, Pennsylvania) covered with transparent tape to diffuse the beam. To obtain LSI images over a wide field of view during in vivo, laboratory measurements, we used a 50-mW, 637-nm laser (Dragon Lasers, ChangChun, China). The maximum optical penetration depth was $\sim 4\mathrm{mm}$ into skin, based on the wavelengths of the lasers used.¹⁰

To facilitate image acquisition, we used custom-written software in MATLAB designed with a touch-sensitive graphical user interface. We attached both the camera and the laser to the back of the tablet via a custom-machined mount.

To ensure that the images were in focus each time the device was repositioned, we oriented the lens of the camera orthogonal to the tablet and the laser pointer at a slight ($\sim 20\mathrm{deg}$) angle with respect to the camera axis. With this alignment, the center of the laser spot moved laterally in the image as the user moved the device closer to or farther from the target. Using a solid reference object as a test target, we verified that, with the laser spot diameter (25 mm) and laser alignment angle, the depth of field (149 mm) of the device was sufficiently large to ensure that the collected raw speckle images remained in focus as long as the laser spot was visible within the field of view.

We converted all raw images into spatial maps of speckle contrast using a $7\times 7$ pixel sliding window and the speckle contrast equation $K=\sigma /⟨I⟩$, where $K$ is speckle contrast and $\sigma$ and $⟨I⟩$ are the standard deviation and mean intensity, respectively, of 49 pixels within the window.¹¹ The position of the sliding window was moved so that the window was centered on each pixel and the corresponding speckle contrast of that pixel calculated with the contrast equation. For each measurement, we collected a set of 250 images at a frame rate of 30 fps. We selected a $100\times 100$ region of interest (ROI) from each of the images and calculated the average speckle contrast from each of the 250 subimages. We then averaged the speckle contrast values to obtain a single average contrast value for each set of 250 subimages.

To facilitate comparison with previously published data,¹² we converted speckle contrast values to maps of speckle flow index (SFI) using the simplified speckle imaging equation $\mathrm{SFI}={(2T{K}^2)}^{-1}$, where $T$ is exposure time and $K$ is speckle contrast.¹³ Previous work from our group suggests a linear response range between SFI and flow speed.¹⁴ SFI maps were used to visualize relative changes in blood flow.

2.3.

Tissue-Simulating Phantoms

To simulate imaging of biological tissues, liquid Intralipid phantoms were synthesized for in vitro validation tests. Liquid phantoms were chosen, as dermal microvascular flow is well represented by Brownian motion, which is mimicked by a well-filled stationary liquid.¹⁵ The speckle contrast of a liquid phantom can be altered by changing the optical properties. LSI is sensitive to both changes in absorption and scattering, due to the effect of optical properties on modulating the pathlength of light. An increase in absorption leads to a suppression of longer pathlength photons, which in turns increases speckle contrast.¹⁶ Intralipid (1%) was added to four 25.4-mm diameter dishes. With the addition of various quantities of India ink by weight, we generated four distinct phantoms with absorption coefficient values over a range (0 to $0.03{\mathrm{mm}}^{-1}$) representative of biological tissues.¹⁷ The phantoms were used for all in vitro experiments.

2.4.

In Vitro Experiments

We acquired images from each of the four phantoms with the device in either a handheld or a mounted configuration. To determine the effect of number of collected images on the variation in speckle contrast values calculated from each image, we acquired three sets of images of one of the tissue-simulating liquid phantoms. The coefficient of variation (COV) of speckle contrast was calculated using the equation $\mathrm{COV}=\sigma /\mu$, where $\sigma$ and $\mu$ are the standard deviation and mean, respectively, of the speckle contrast values for individual subsets of consecutive speckle contrast images. The number of images included in the COV calculation served as the independent variable.

To study the effect of motion artifacts associated with handheld operation of the LSI device, we imaged the tissue-simulating liquid phantoms at different camera exposure times. To compare the handheld data with the mounted data, we used logarithmic regression and performed a linear fit to the mounted data. We then calculated the sum of squares of the fit to the data and calculated the quality of fit for each phantom measurement.

We then investigated the repeatability of measurements across multiple users. Seven volunteers used the handheld device to image the same four liquid phantoms described above. The exposure time was set at 5 ms and the lens aperture at $f/22$. Each user was instructed on how to use the device and subsequently collected three sets of images from each phantom. After each user acquired images from each of the four phantoms, we mounted the device in a fixed position and collected an identical number of images. We used a two-way ANOVA test to assess if the group mean for each phantom measurement differed significantly ($p<0.05$) depending on (1) the concentration of India ink, (2) the imaging configuration, and (3) any interaction between the concentration and configuration.

2.5.

In Vivo Experiments

Two occlusion models were used to assess the qualitative and quantitative performance of the device to identify regions of altered blood flow. As a qualitative demonstration, we induced occlusion of the fingertip by tightly wrapping a rubber band around the knuckle of the finger of a volunteer. We illuminated the entire hand to obtain a wide-field blood-flow map in an attempt to identify a region of decreased blood flow to the occluded digit.

To demonstrate the ability of the device to provide a quantitative assessment of in vivo blood-flow dynamics, we performed a reactive hyperemia experiment⁸ on one human subject. We used a sphygmomanometer to apply a pressure of 180 mmHg for 1 min to temporarily occlude the brachial artery. We held the LSI device to collect raw speckle images from the palm of the subject. We then released the applied pressure and continued to collect raw speckle images to visualize the known hyperemic response to the occlusion challenge. We then selected an ROI and calculated the average SFI value for each frame.

2.6.

Clinical Case Study: Neonatal Intensive Care Unit

In an ongoing feasibility study, we used the handheld LSI device to collect data from a neonate, which developed a clot in the right forearm that was confirmed with Doppler ultrasound. Tissue downstream of the clot became hypoxic due to the blockage of flow. The peripheral vasculature in the thumb was most affected and led to irreversible tissue damage. Handheld LSI measurements were taken of the affected hand to determine if a perfusion deficit could be identified in the thumb.

3.1.

In Vitro Experiments

In general, the COV decreased with an increase in the number of frames that were averaged (Fig. 2). We postulate that the large fluctuations in the COV function when fewer frames are averaged are due to the distribution of the speckle contrast values approaching normality as the number of frames increased. The COV was less than 2% after averaging over 160 frames, which corresponds to an image acquisition period of 5.3 s.

Fig. 2

COV calculated as a function of the number of frames averaged. The COV was calculated from each set of images ($n=3$). The COV decreased below 2% with analysis of 160 frames collected over 5.3 s.

3.1.1.

Single, experienced user

For each concentration of India ink, we observed a decrease in speckle contrast with an increase in exposure time (Fig. 3). The trend was similar between the data collected in a handheld and mounted configuration. Based on the $R^2$ coefficients (0.72 to 0.87) calculated from comparison of the handheld and mounted data, we conclude that, with a single, experienced user, handheld motion artifacts had a negligible effect on the accuracy of the data collected with the device.

Fig. 3

Comparison of single-user handheld measurements of speckle contrast with those obtained using a mounted device. Over a range of absorption coefficients (varied with India ink fraction), speckle contrast values were similar with both configurations. Error bars represent the standard deviation of the mean contrast value ($n=4$ for each data point). The $R^2$ values represent the goodness of fit associated with comparison of the log of the handheld and mounted contrast values. Lines are used strictly as a visual aid for the reader and do not represent model-based fitting of the data.

3.1.2.

Multiple users

With operation by multiple users, we did not identify any significant difference ($p>0.05$) between the measured speckle contrast values using either a handheld or mounted configuration (Fig. 4). We also did not identify any significant effect ($p>0.05$) from the interactions between the configuration used during acquisition and the concentration of India ink. Finally, we identified a significant difference ($p<0.05$) between the measured speckle contrast values of each concentration of India ink. Collectively, these data provide support of the ability of multiple operators to use handheld LSI to measure speckle contrast values that are similar to those measured with the conventional mounted configuration.

Fig. 4

Comparison of speckle contrast values from handheld measurements taken by multiple users ($n=7$). Measurements collected with a mounted device are included for comparison. The mean contrast value was taken from each data set ($n=3$ sets per user). The measured speckle contrast values were similar with both configurations. Error bars represent the standard deviation of the mean contrast value ($n=21$ for each bar).

3.2.

In Vivo Experiments

3.2.1.

Occlusion models

With the fingertip occlusion model, we observed a decrease in SFI in the occluded fingertip compared with values in the surrounding regions of the hand (Fig. 5). The decrease in SFI coincides with the expected decrease in blood flow associated with the occlusion. These data support the ability of the handheld LSI device to qualitatively detect regions of compromised blood flow in a laboratory.

Fig. 5

SFI image of a hand during fingertip occlusion with a rubber band. We collected the raw speckle images using the LSI device in handheld mode. The occluded fingertip had a lower SFI value than the other fingertips.

With the reactive hyperemia model, a decrease in SFI was observed, coinciding with the occlusion of the brachial artery from $t=20$ to 95 s. The SFI values initially increased and returned to baseline after the occlusive pressure was released. These changes in SFI are in agreement with the expected trend of hyperemic blood flow following cuff release⁸ (Fig. 6). These data support the ability of handheld LSI to reliably image quantitative blood-flow dynamics from human subjects.

Fig. 6

Average SFI versus time measured from a human palm during brachial artery occlusion and release. A pressure of 180 mmHg was applied at the 20-s mark and maintained until the 95-s mark. The applied pressure was then released, resulting in a hyperemic response. A moving average filter was applied to the average signal for clarity (in red).

3.2.2.

Clinical case study: neonatal intensive care unit

With handheld LSI of the hand of a neonate, we readily identified a region of compromised blood flow in the digit occluded due to a confirmed blood clot in the forearm (Fig. 7). These data suggest a potential role of handheld LSI in a challenging clinical environment such as the neonatal intensive care unit, to enable routine measurements of blood flow in regions of interest, potentially by nonexperts (i.e., clinical staff).

Fig. 7

Blood-flow map of the right hand of a neonate (53 days postgestation) in the neonatal intensive care unit at UC-Irvine Medical Clinic. The thumb (located inside of the red rectangle) of the subject had a clear reduction in blood flow, resulting from a clot in the right forearm. Red = highest perfusion, dark blue = lowest perfusion. Note that an outline (in black) of the hand and fingers was superimposed for clarity.