2.1.

Retinal Imaging

Retinal imaging was performed with a manufacturer-modified Heidelberg Retina Tomograph (HRT)-classic (Heidelberg Engineering, Heidelberg, Germany), which is a confocal SLO. The modified HRT-classic could originally image with a 20, 15 or 10 deg field of view (FOV), but it was modified such that it could image at smaller FOVs of 10, 2, or 1 deg. Thus, it is capable of very high-resolution imaging of the retina. Images taken at each FOV of the modified HRT-classic can be seen in Fig. 1. Images are $256\times 256\text{pixels}$. Ocular refraction can affect image magnification, but each pixel subtends $\sim 10\mu \text{m}$ at 10 deg FOV, 2 μm at 2 deg FOV, and 1 μm at 1 deg FOV. For the assessment of the reactivity, response to hyperoxia subjects were imaged at 2 deg FOV.

Fig. 1

Images of the same portion of the retina taken at different fields of view (FOV) of the modified HRT-classic. (a) 10 deg FOV, (b) 2 deg FOV, and (c) 1 deg FOV.

The modified HRT-classic images using a 675-nm diode laser. During image acquisition, at any given instant, only one point on the retina is illuminated. Therefore, to acquire a two-dimensional image, the emitted light beam must be swept over the retina. The emitted beam is scanned in the $x$ and $y$ axes along the plane of focus perpendicular to the optic axis ($z$ axis) using two oscillating mirrors.²⁷ In this way, an image is recorded point-by-point, over time, as the region of interest is scanned in a raster-like fashion. The fast scan is in the horizontal direction, with a scanning rate of 8 kHz, while the vertical scan rate is 30 Hz. It scans in double-scan mode, which means that the first line is recorded as the scanner moves from left to right and the second line is acquired as the beam scans back from right to left. Thus, two lines of an image are obtained in one complete cycle of the fast horizontal scanner.²⁷

This system has optical sectioning capabilities; however, image sequences of 32 frames were acquired at a single focal plane to obtain a video of the vasculature. The acquisition time for a single frame was 32 ms, while the acquisition time for a sequence of 32 frames was 1.5 s, with a resetting time of 16 ms between frames.

The fixation target used was a Samsung Galaxy Tab™ (Seoul, Republic of Korea), which displayed a red dot that could be altered in size and moved around the screen by the investigator to center the retinal region of interest in the captured image. Subjects fixated on the target with the eye opposite to the one being imaged.

2.2.

Desinusoiding and Image Registration

Raw videos were first processed to correct for distortions due to raster scanning. The SLO employs a sensor that captures pixels at a constant rate and a resonant scanner whose velocity varies sinusoidally across each scan line. This results in a horizontal distortion in raw videos. The velocity of the scanner is slowest at the edges of the frame and fastest in the center; accordingly, more pixels are recorded per retinal area toward the edges compared to the center of the image. Videos of a calibration grid were used to estimate this horizontal warping, and a desinusoiding algorithm was developed in MATLAB® to correct for this distortion. Briefly, the algorithm used a cost function, whose value was high when the lines of the calibration grid in the image were irregularly spaced and low (close to zero) when evenly spaced, and the Nelder-Mead simplex search algorithm to calculate the transformation necessary to dewarp the images.

The next step was image stabilization to correct for eye motion to allow for image averaging to produce a high signal-to-noise retinal image. The eye is constantly moving, even during attempts to fixate steadily on a stationary target. All retinal imaging systems suffer from these eye movements; however, these distortions are particularly apparent in adaptive optics ophthalmic instruments and the modified HRT-classic, as the distortion scales with the magnification of the instrument.^28–30 Eye motion that occurs during the acquisition of image sequences acquired by scanning techniques produces distortions that are unique in each frame because the images are captured over time from a scanning focused beam. These distortions can be extreme if a microsaccade occurs during acquisition, as can be seen in Fig. 2. Therefore, registration of retinal images acquired using scanning techniques is more complex than implementing a simple rigid translation to align the images. However, during short time intervals, retinal motion can be well approximated by a simple translation. (Note that if the images have not been properly desinusoided, this assumption is not valid—desinusoiding is essential before registration.) This has led to patch-based correlation methods being used to register small-FOV SLO images.^28,29,31 We implemented such a patch-based registration scheme to align our image sequences. Because the direction of the fast scan in the SLO is horizontal, it is assumed that a single horizontal scan line is unaffected by fixational eye motion and is essentially a snapshot of the retina. Ideally, each patch would have extended just one line vertically (the direction of the slow scanner); however, in most sequences, such patches fail to register due to noise. A compromise between having enough data to reliably match the patches and obtaining the most accurate within-frame motion estimates must be reached. Thus, each template image was partitioned into patches that extend horizontally across the entire image but extend to just a few lines of the vertical extent of the image.

Fig. 2

Example of extreme distortion that can occur when using scanning techniques. Three sequential images taken from a sequence obtained by the modified HRT-classic are shown. The second image shows severe distortion due to a microsaccade, which occurred during image acquisition.

The correlation coefficient computed in the frequency domain, as described in a paper by Lewis,³² was the match metric used in the MATLAB®-based algorithm to register sequences of images from the HRT. It is necessary to use the correlation coefficient to make shift estimations robust to intensity fluctuations. Such intensity fluctuations are very common in image sequences obtained from ophthalmoscopes due to subject misalignment, tear film break-up, etc. Another useful property of the correlation coefficient is that it has an absolute range [$-1$, 1]. This was used to delete frames that did not sufficiently match the reference image by discarding images whose maximum correlation coefficient did not reach a set threshold. This could be due to insufficient overlap between the reference and template images, which occurred when the eye wandered too far away from the fixation target, or due to extreme warping in the template image. Before importing the desinusoided image sequence into the registration algorithm, frames where blinks had occurred, resulting in the intensity dropping to zero in the images throughout the blink, were manually deleted.

The first step in our registration algorithm was to implement a full-image alignment. The images underwent a full-image registration, before applying the patch-based scheme, as our images were small, extending just 256 pixels horizontally, and therefore, removing the interframe eye-motion made the patch-based, intraframe registration more robust. The first image in a sequence was normally taken as the reference image, unless it showed visible signs of distortion. The same image was used as the reference image in the full-image and patch-based registration steps.

Once the images were registered, high-signal-to-noise images were obtained by averaging the sequence of images. Figure 3 shows single frames and temporal averages of registered sequences. An improvement in signal-to-noise in the average image, Fig. 3(b), compared to the single image, Fig. 3(a), is evidence of the registration accuracy. In Fig. 3(c), noise was not as prevalent and individual photoreceptors can be distinguished; the corresponding average image, Fig. 3(d), shows no loss of detail, which would occur if the images were not accurately registered.

Fig. 3

Left: (a) and (c) are single frames of image sequences obtained from the modified HRT-classic. Right: (b) and (d) are the temporal averages of the image sequences after registration. In both of these examples, all 32 frames of the image sequences were retained after image registration and used in the image averaging step.

2.3.

Vessel Tracking and Diameter Measurement

A semiautomated vessel tracking and diameter measurement method was implemented. This method was based on the assumption that an arteriole can be considered as a curved cylinder consisting of a linkage of many small, straight cylinders. The MATLAB®-based algorithm commenced by importing an averaged image of an arteriolar segment. A low-pass Gaussian filter was applied to the average image to reduce sharp transitions in background luminance. The smoothed image was then displayed and the user selected $n$ seed points, ${\mathrm{sp}}_n$, at the center of the vessel describing the path of the vessel. The seed points were selected such that the section of vessel between two successive seed points was approximately straight. Using these seed points, the vessel was then divided into $n-1$ segments, where ${\text{segment}}_n$ consisted of the interval $[{\mathrm{sp}}_n,{\mathrm{sp}}_{n+1}]$. The direction of each segment, ${\alpha }_n$, was calculated. Estimated center points, cp, were calculated along the segment. Estimated center points were calculated to be 2 pixels apart in the segment direction ${\alpha }_n$. A cross-section line was calculated at each center point, perpendicular to the segment direction and centered at the center point, Fig. 4(a). A vessel profile, Fig. 4(b)(i), was then extracted at each cross-section line.

Fig. 4

(a) Estimated center points and cross-section lines. (b) The sliding linear regression filter method was applied to the vessel profiles (i) and the center and edge points were determined from the resulting gradient data (ii). (c) The center and edge points resulting from using this technique.

The vessel edges and center positions were identified using a sliding linear regression filter method proposed by Chapman et al.³³ This technique is based on fitting a best fit line by linear regression relating image intensity against distance along the profile within a sliding window filter. The window of $W$ points centered on the $i$’th point was progressively moved point-by-point along the profile, as demonstrated in Fig. 4(b)(i). At each point, a straight line was fit to the intensity values within the window using the least-squares method. The gradient of the straight line was recorded at each point, $m_i$. Figure 4(b)(ii) shows the corresponding gradient data of the vessel profile. The edge points were determined from this gradient data. The left edge was taken as the minimum slope obtained from the left half of the profile. The right edge was taken as the maximum slope obtained for the right half of the profile. The actual center points of the vessel were found by taking the central portion of the vessel profile and locating the point at which the gradient changed from positive to negative. This process is demonstrated in Fig. 4(b). Vessel diameters were then set as the distance between corresponding left and right edges, giving a diameter value for each center point. The median of these vessel widths was taken as the vessel diameter.

In this method, one needs to consider the window size $W$. Large window sizes reduce the occurrence of vessel edges being incorrectly located at random bright/dark spots in the background, while short windows provide better edge resolution, especially for small vessels, but at the expense of reduced noise rejection. For this reason, this technique tracked the vessel twice to incorporate adaptation of the size of the filter window to the approximate size of the vessel. During the first time of tracking, an estimate of the vessel edges, center points, and diameters was calculated using a large cross-section line (100 pixels) and a long window length (10 pixels). For the second time of tracking, the length of the cross-section line was set to be only slightly greater than the median diameter, which was estimated in the first run, and a shorter filter was applied ($\sim 10\%$ of the length of the cross-section line). Therefore, during the second run, despite the reduced noise rejection due to the shorter filter length, the edge points were rarely identified at random points in the background. This was because little background intensity was included in the extracted profile due to the cross-section line being adapted to an estimate of the vessel diameter, so there were few random bright/dark spots that could mistakenly have been determined as edge points. The diameter of the arteriole was set as the median diameter calculated from the second time of tracking.

In Fig. 4(c), the edge points displayed were calculated from the resulting arteriole diameter, centered at the calculated center points and perpendicular to the vessel direction. From this, it can be seen that this method delineates the vessel well and provides a good estimate of the arteriole diameter.

2.4.

Retinal Mosaics

At small FOVs, it can be difficult to locate vessels of interest and repeatedly return to them for multiple measurements. Therefore, approximately nine 10 deg FOV (the largest FOV of the modified HRT-classic) image sequences were obtained from one eye of each participant and were used to create a mosaic of the retina. An example of one such mosaic can be seen in Fig. 5. The mosaic was used during the reactivity measurements to define suitable measurements sites (relatively straight arteriole segments with good contrast to the background) and to aid the investigator in navigating the retina at a small FOV. Having the mosaic essentially provided the investigator with a map of the retinal vasculature, greatly increasing the speed and efficiency of imaging at high resolution. The mosaic of the 10 deg images was generated by first desinusoiding the images and then combining overlapping images using Adobe Photoshop®.

Fig. 5

A typical mosaic, composed of nine 10 deg FOV images.

2.5.

Oxygen Delivery System

Subjects were administered 100% oxygen via a nonrebreathing system at a flow rate of $12\mathrm{L}/\min$ at atmospheric pressure for $\sim 20\min$. The nonrebreathing system consisted of an oxygen cylinder, regulator and flowmeter (which adjust the oxygen flow), and a nonrebreathing mask. A nonrebreathing system ensures a high flow of oxygen and a minimal intake of room air, resulting in an inspired oxygen concentration of 80 to 90%. The subjects were instructed to report any noticed effects of oxygen inhalation; none of the volunteers noted any adverse reaction to oxygen breathing.

3.1.

Subjects

Twenty-nine (6 women and 23 men) healthy, normal volunteers (mean $\text{age}\pm \mathrm{SD}$, $22.1\pm 3.7\text{years}$) participated in this study, including 17 (4 women and 13 men) nonsmokers (mean $\text{age}\pm \mathrm{SD}$, $21.9\pm 2.6\text{years}$) and 12 (2 women and 10 men) smokers (mean $\text{age}\pm \mathrm{SD}$, $22.4\pm 5.1\text{years}$). Each subject signed a written consent form after the nature of the study and experiment protocol were explained in detail. All procedures conformed to the tenets of the Declaration of Helsinki and were reviewed and approved by the National University of Ireland, Galway Ethics Committee. Exclusion criteria included ages $<18\text{years}$ or $>40\text{years}$, unclear optic media, pregnant or breastfeeding women, any cardiovascular or respiratory disorders, taking medication with known effects on blood flow (e.g., muscle relaxants), and taking any chronic medication. Smoking status was identified by self-report. The average number of cigarettes smoked by the smokers per day ranged from 3 to 20 ($\text{mean}\pm \mathrm{SD}$, $10.9\pm 6.3$ cigarettes), and the duration of smoking varied from 1 to 20 years ($\text{mean}\pm \mathrm{SD}$, $6.1\pm 5.7$).

3.2.

Experimental Protocol

Volunteers attended three sessions. The first session was to determine eligibility, to familiarize the subjects with the measurement technique, and to obtain some large-FOV images. These images were used to create the retinal mosaics. One eye from each subject was selected. The magnitude of retinal vascular reactivity in response to the hyperoxic stimulus was measured during the second and third sessions, with intervals of 1 to 17 days between these sessions. Subjects were asked to abstain from alcohol, caffeine, and cigarette smoking (in the case of smokers) for 12 h before study sessions 2 and 3. Subjects rested for at least 5 min before the start of each experiment to stabilize baseline cardiovascular and respiratory parameters. All testing was performed in a quiet laboratory, at normal room temperature, and at sea-level atmospheric pressure.

In the nonsmoking group, session 2 was performed to assess the reaction and recovery times of arterioles and session 3 was to ascertain if regional differences in retinal vascular reactivity existed. Therefore, the experimental protocol was slightly different for sessions 2 (protocol A) and 3 (protocol B).

In the smoking group, protocol A was used for both sessions 2 and 3. Session 2 was performed to assess the long-term effect of smoking on the magnitude of constriction of arterioles in response to hyperoxia and session 3 was performed to investigate the acute effect of smoking on the reactivity response. For session 2, the experimental protocol was performed after a 12-h abstinence from cigarette smoking. Volunteers also refrained from smoking for 12 h before session 3; however, they were asked to smoke a single cigarette just before the start of the experiment. The test protocol was otherwise identical for both sessions for the smokers. To assess the effect of smoking on the parameters measured, data were compared to the data obtained for the nonsmokers. As there was no significant difference between any parameters obtained for nonsmokers between sessions 2 and 3, the nonsmokers’ data were averaged across both sessions when being compared to the smokers’ data.

To assess the relationship between the magnitude of the reactivity response to hyperoxia and vessel caliber, data from image sequences from the nonsmoking group, which contained both a large and small arteriole, were collated. Arterioles with diameters $>30\text{pixels}$ ($\sim 60\mu \text{m}$) were categorized as large, while those with diameters $<30\text{pixels}$ were classed as small.

Blood pressure was measured during sessions 2 and 3. Three measurements were taken during each phase. Systolic (${\mathrm{BP}}_{\mathrm{s}}$) and diastolic blood pressures (${\mathrm{BP}}_{\mathrm{d}}$) were measured on the upper arm by an automated oscillometric device (Omron M10-IT; Omron Healthcare Europe BV; Hoofddorp, The Netherlands). Mean brachial artery blood pressure (${\mathrm{BP}}_{\mathrm{m}}$) was calculated by ${\mathrm{BP}}_{\mathrm{m}}={\mathrm{BP}}_{\mathrm{d}}+(1/3)({\mathrm{BP}}_{\mathrm{s}}-{\mathrm{BP}}_{\mathrm{d}})$. Pulse rate (PR) and oxygen saturation (${\mathrm{SpO}}_2$) were also recorded during both reactivity measurement sessions. Three readings were taken during each phase from a finger pulse oximetric device (Ohmeda Biox 3700e; Datex-Ohmeda Inc., Madison, Wisconsin). All measurements were made by the same investigator.

Diameter measurements were not converted from pixels to micrometers using axial length measurements because relative change was the primary outcome measure desired. However, for the 2 deg FOV, 1 pixel is approximately equal to 2 μm.

A change in IOP causes a change in retinal perfusion pressure. Previous studies have measured IOP at baseline and during oxygen breathing and found that IOP does not change significantly during hyperoxia.^18,34–36 Previous studies have also investigated the effect of cigarette smoking on IOP. IOP was measured before and after cigarette smoking and it was found to be unaltered by acute smoking.^{20,21,37–39} It has also been established that there is no difference in IOP values between smokers and nonsmokers.^9,22,38 Consequently, IOP was not recorded during our experiments.

Previous studies have demonstrated that the magnitude of retinal arteriolar reactivity is independent of gender^7,14,22,34 and left/right eye;^7,22 we, therefore, did not investigate reactivity as a function of these parameters.

3.3.

Data Analysis

Statistical analysis was performed using Minitab® software package (Release 16; Minitab Inc., State College, Pennsylvania). Each dataset was group-averaged across subjects as a function of phase (i.e., baseline, hyperoxia, recovery). The response of arteriole diameters to hyperoxia is expressed as a percentage change as compared to baseline. Datasets were assessed to determine if they were normally distributed using the Anderson-Darling test. To investigate the influence of oxygen breathing on parameters, two-tailed paired $t$ tests were performed between baseline values and hyperoxic values. To compare the effect of hyperoxia between two different groups (e.g., nonsmokers versus smokers, or large versus small arterioles), a two-tailed independent samples $t$ test was used. A comparison of more than two groups (e.g., comparing across retinal quadrants) was performed using one-way analysis of variance (ANOVA). Variability of measurements was assessed by calculating the standard error and by comparing the results obtained between sessions 2 and 3 in the nonsmoking group. All data are presented as $\text{mean}\pm \text{standard error}$ (SE). The level of significance was set to $p<0.05$.

4.1.

Young, Healthy Nonsmokers

The group-averaged values for each systemic variable for each phase of session 2 are presented in Table 1, and the values obtained for each phase of session 3 are presented in Table 2.

Table 1

Systemic hemodynamics obtained throughout session 2. Data are presented as mean±SE. t tests were not performed on SpO2 data as they were not found to be normally distributed.

Table 2

Systemic hemodynamics obtained throughout session 3. Data are presented as mean±SE. t tests were not performed on SpO2 data as they were not found to be normally distributed.

The results from session 2 provided us with diameter measurements throughout the experiment, allowing us to plot the diameter changes during the different phases of the experiment. The time course of the response of an arteriole diameter of a single, typical participant to a hyperoxic provocation as a function of time is shown in Fig. 6. It can be seen that vasoconstriction occurred during the first 5 min of oxygen breathing; thereafter, the diameter remained stable for the remaining 15 min of oxygen breathing. Upon cessation of the hyperoxic phase, the arteriolar diameter began to increase back to baseline. It is apparent from the plot that the reaction of the arteriole diameter to the hyperoxic stimulus is faster than the recovery back to baseline. After $\sim 15\min$ of breathing room air, at the end of the oxygen breathing phase, the arteriole had returned to its prehyperoxia testing diameter. The group mean retinal arteriole diameter was $39.6\pm 1.5\text{pixels}$ at baseline. This decreased significantly to $36.5\pm 1.3\text{pixels}$ during oxygen breathing ($p=0.001$) and recovered to $39.1\pm 1.6\text{pixels}$ during the recovery phase of room-air breathing. The group mean percentage change in arteriolar diameter during hyperoxic provocation was $-9.7\pm 1.0\%$.

Fig. 6

Change in the retinal arteriole diameter of a single, typical participant induced by hyperoxia as a function of time. Vertical lines represent the points at which oxygen breathing started and stopped. A pixel subtends $\sim 2\mu \text{m}$. The arteriole diameter decreases during the first 5 min of oxygen breathing, plateaus, and remains stable for the rest of the oxygen breathing phase. The arteriole diameter increases during the air breathing, recovery phase and has returned to baseline after $\sim 15\min$.

Figure 7(a) presents a comparison of the group mean, baseline arteriolar diameters in the superior temporal (ST), inferior temporal (IT), superior nasal (SN), and inferior nasal (IN) retinal quadrants. Table 3 details the group-averaged arteriole diameter values across quadrants and hemiretinae (i.e., temporal, nasal, superior, inferior). It is apparent that the temporal arterioles are larger than the nasal arterioles ($p=0.008$), while superior arterioles do not differ from inferior arterioles ($p=0.725$). It is also evident that significant vasoconstriction occurred in all retinal regions. Figure 7(b) presents a comparison of the group-averaged percentage change in the arteriole diameters, relative to baseline, across the retinal quadrants during hyperoxia. Table 4 details the percentage change in diameter obtained for each retinal region. The data indicate that systemic hyperoxia induced a group-averaged arteriole constriction of $-9.2\pm 0.9\%$ (ST), $-9.0\pm 0.9\%$ (IT), $-11.0\pm 1.2\%$ (SN), and $-10.6\pm 0.9\%$ (IN). The degree of constriction does not vary across the quadrants ($p=0.372$, ANOVA). A comparison of the magnitude of arteriolar constriction in the hemiretinal regions, i.e., temporal versus nasal and superior versus inferior, revealed that the arterioles in the temporal retina appeared to constrict slightly less than those nasally, but this was not significant ($p=0.080$), and the arterioles in the superior retina responded to the same degree as those in the inferior retina ($p=0.725$).

Fig. 7

(a) Baseline, group mean arteriole diameters in the retinal quadrants. A pixel subtends $\sim 2\mu \text{m}$. (b) Comparison of the percentage change in arteriolar diameter from baseline, across retinal quadrants. Data are presented as $\text{mean}\pm \mathrm{SE}$.

Table 3

Comparing arteriole diameters across retinal regions, before and during hyperoxic provocation. Data are presented as mean±SE.

Table 4

Comparing the percentage difference of arteriole diameters due to hyperoxia, compared to baseline, across retinal regions. Data are presented as mean±SE.

The group mean baseline diameter for the narrow arteriolar measurement site was $19.2\pm 0.8\text{pixels}$ ($\sim 38.4\pm 1.6\mu \text{m}$) ($n=18$), and for the wide measurement site, it was $36.1\pm 0.7\text{pixels}$ ($72.2\pm 1.4\mu \text{m}$) ($n=22$). The group mean diameter constriction of the small arterioles was $-1.9\pm 0.2\text{pixels}$ and for the large arterioles, it was $-3.4\pm 0.2\text{pixels}$. There was a significant difference in the absolute change in diameter between large and small arterioles ($p=0.000$). The group-averaged percentage change in diameter of the small arterioles was $-9.8\pm 0.8\%$ and for large arterioles, it was $-9.2\pm 0.6\%$. There was no difference in the percentage constriction between the large and small vessels ($p=0.538$).

The arteriole imaged during session 2, to study the reaction and recovery time, was also chosen as one of the four vessels imaged in session 3, to study regional differences in reactivity. To assess the repeatability of our measurements, the values obtained for these arterioles during the two sessions are compared. The group mean baseline diameter of the arterioles in session 2 was $38.1\pm 2.1\text{pixels}$ and in session 3, it was found to be $39.1\pm 1.9\text{pixels}$. There was no difference in the baseline diameters between visits ($p=0.907$). During the hyperoxic phase of session 2, the arterioles decreased to $35.2\pm 1.9\text{pixels}$, and in session 3, they decreased to $35.6\pm 1.9\text{pixels}$. There was no difference, between the two visits, in the hyperoxic diameters measured ($p=0.888$). The group-averaged constriction reached in session 2 was $-9.3\pm 1.2\%$ and in session 3, it was found to be $-9.6\pm 1.2\%$. There was no difference in the magnitude of vasoconstriction between visits ($p=0.871$), while the mean difference in the percentage change of diameter between visits was $0.3\pm 0.8\%$ ($\text{mean}\pm \mathrm{SE}$).

4.2.

Young, Otherwise Healthy Smokers

The group-averaged values for each systemic variable for each phase of session 2 (abstinence from smoking for 12 h, ${\text{smokers}}_{[\mathrm{no}\mathrm{cig}]}$) are presented in Table 5, and the values obtained for each phase of session 3 (cigarette smoked immediately before commencement of experiment, ${\text{smokers}}_{[\mathrm{post}\mathrm{cig}]}$) are presented in Table 6.

Table 5

Systemic hemodynamics obtained throughout session 2 (no cigarette). Data are presented as mean±SE. t tests were not performed on SpO2 data as they were not found to be normally distributed.

Table 6

Systemic hemodynamics obtained throughout session 3 (post cigarette). Data are presented as mean±SE. t tests were not performed on SpO2 data as they were not found to be normally distributed.

Table 7 details the group-averaged arteriole diameter values across groups (i.e., ${\text{smokers}}_{[\mathrm{no}\mathrm{cig}]}$, ${\text{smokers}}_{[\mathrm{post}\mathrm{cig}]}$, and nonsmokers). Figure 8(a) presents a comparison of the group mean, baseline arteriolar diameters in ${\text{smokers}}_{[\mathrm{no}\mathrm{cig}]}$, ${\text{smokers}}_{[\mathrm{post}\mathrm{cig}]}$, and nonsmokers. It is evident from this plot that the baseline diameters did not differ between ${\text{smokers}}_{[\mathrm{no}\mathrm{cig}]}$ and ${\text{smokers}}_{[\mathrm{post}\mathrm{cig}]}$. Although the baseline group mean arteriole diameters of the nonsmoking group appear to be a little larger than those of the smoking group, we did not find a significant difference between the groups ($p=0.186$). It is also apparent from the data in Table 7 that significant vasoconstriction occurred in all three groups.

Table 7

Comparing arteriole diameters for each experimental phase (i.e., baseline, hyperoxia, and recovery) by smoking status. Data are presented as mean±SE.

Fig. 8

(a) Baseline, group mean arteriolar diameters in the retinal quadrants. A pixel subtends $\sim 2\mu \text{m}$. (b) Comparison of the percentage change in arteriolar diameter from baseline, across retinal quadrants. Data are presented as $\text{mean}\pm \mathrm{SE}$.

Table 8 details the percentage change in diameter obtained for each group. Figure 8(b) presents a comparison of the group-averaged percentage change in the arteriole diameters, relative to baseline, attained by ${\text{smokers}}_{[\mathrm{no}\mathrm{cig}]}$, ${\text{smokers}}_{[\mathrm{post}\mathrm{cig}]}$, and nonsmokers, during the hyperoxic phase. The data indicate that the degree of vasoconstriction achieved by the smokers was the same when measured after a 12-h abstinence and immediately after smoking a cigarette ($p=0.621$). The reactivity of smokers was, however, significantly less than that of nonsmokers (${\text{smokers}}_{[\mathrm{no}\mathrm{cig}]}$ versus nonsmokers, $p=0.017$; ${\text{smokers}}_{[\mathrm{post}\mathrm{cig}]}$ versus nonsmokers, $p=0.004$). In the recovery phase, the arteriolar diameters of each group returned to their prehyperoxic diameters.

Table 8

Comparing the percentage difference in arteriole diameters due to hyperoxia, compared to baseline, across smoking status. Data are presented as mean±SE.

The time course of the response of an arteriole diameter of a single, typical smoker to hyperoxic provocation as a function of time is portrayed in Fig. 9. The red dots represent the response of the arteriole diameter to hyperoxia after a 12-h abstinence from smoking. The black crosses describe the response of the diameter of the same arteriole segment to hyperoxia immediately after smoking a cigarette. The vertical lines in the plot mark the points at which oxygen breathing started and stopped. It is evident that smoking a cigarette had no effect on the response of the arteriole diameter to hyperoxia.

Fig. 9

Change in the retinal arteriole diameter of a single, typical smoker, induced by hyperoxia as a function of time. The red dots depict the response to the arteriole diameter to hyperoxia after a 12-h abstinence from smoking. The black crosses delineate the arteriolar diameter response immediately after smoking a cigarette. The vertical lines mark the points at which oxygen breathing started and stopped. There is no difference in the arteriolar response to hyperoxia between sessions.