Blood vessel walls consist of three main layers, the tunica intima, tunica media and tunica adventitia. Hypertension has different impacts on each of these three layers. The tunica intima is mainly composed of endothelium and it thickens after exposure to high blood pressure, mainly due to accumulation of extracellular matrix materials such as collagen 27. The middle layer – the tunica media - is composed of muscular tissue embedded in an extracellular matrix of elastin fibers and this layer is birefringent 20. With hypertension, the tunica media undergoes degenerative changes such as fibrinoid necrosis (cell death), destruction of elastin fibers and muscular remodeling (thickening of the muscles) 28. The outer most layer of the blood vessels, the tunica adventitia, is made of connective tissue and an elastic membrane. During hypertension, degradation occurs in the connective tissue 29. Therefore, changes in retinal blood vessels associated with hypertension are listed under three categories:
-
Classic vascular or non-vascular changes (attributed to hypertensive retinopathy) in the retina is based on prognosis of different retinal signs in response to high blood pressure. These signs that have become the main subject for traditional hypertensive retinopathy classification such as Keith-Wagener-Barker (KWB) 30 and the Mitchell-Wong 31 classification systems include optic disc swelling, arteriolar narrowing or nicking, retinal haemorrhages, exudates, cotton-wool spots, copper- and silver-wire arteriole appearance and macular edema which are usually extracted from scanning laser ophthalmology (SLO) or simply by assessing the retinal fundus images. Although these classification systems are generally used in clinics, they do not provide any information on the internal blood vessel wall structure.
-
Changes in the caliber of the retinal blood vessels using parameters such as wall to lumen ratio (WLR), media thickness to internal lumen ratio (MLR) and arteriovenous ratio (AVR) are extracted using retinal images taken with adaptive-optics-based cameras (AO-OCT/AO-SLO) or scanning laser Doppler flowmetry (SLDF) combined with computerized methods or invasive micromyography 14,32,33.
-
Evaluation of the global geometry of the retinal blood vessels by implementing artificial intelligence (AI)-assisted methods on fundus images for assessment of retinal vasculature branching parameters that may indicate risk of vascular damage based on deviations in blood vessel tortuosity, fractal dimension, branching angles, and ratio of vascular length to diameter. However, these deviations have not yet been demonstrated to correlate with CVDs 34.
In this study we assessed retinal blood vessels adjacent to the optic nerve head in hypertensive and normotensive subjects to find differences in the optical properties of the vessel wall tissue using PS-OCT-based BBI measurements. This new method does not belong to any of the three categories mentioned earlier, as BBI is a measure for the structural integrity of the retinal blood vessels walls. Impacts of hypertension on different layers of the retinal blood vessels are mentioned in the supplementary information.
In our hypertensive subjects, retinal blood vessel walls had lower DPPR/UD values compared to healthy age-matched subjects (Fig. 1A and 1B). In addition, our data indicated that retinal arteries were affected more by hypertension in comparison to veins (Fig. 1B and 1C). Mean DPPR/UD reduction in artery and vein walls were 27% and 20%, respectively. We speculate this is caused by the higher pressure and blood flow rate in arteries compared to veins 35. Our results also demonstrated that both arteries and veins experienced an increase in thickness of about 9 ± 3 µm, on average, due to hypertension. The thickening of the blood vessel walls in response to higher blood pressures is well documented 36,37. Blood vessels respond to a change in the blood pressure via wall mass reinforcement 38. However, our wall thickness results are not in agreement with other studies, emphasizing the importance of a birefringence assessment for accurate thickness measurements and the detection of hypertension. For example, Muraoka et al. used OCT intensity images and showed that mean arterial wall thickness of the normotensive subjects increased slightly from 17.7 µm to 18.3 µm in hypertensive subjects. The mean venous thickness increased from 13.9 µm to 14.5 µm 16. While their study did not aim to distinguish between healthy and unhealthy blood vessels, these small differences in thickness were not specific and sensitive enough to be used as a diagnostic disease classifier.
In addition, as we mentioned in our previous study 20, results of the studies that determine the thickness of the blood vessel walls based on OCT images do not always agree with each other. The discrepancy in different studies regarding the retinal blood vessel walls thickness stems mostly from lack of contrast in OCT intensity images and consequently incorrect determination of blood vessel wall boundaries. Hence, it is not practical to rely on thickness measurements based on intensity OCT images. Without access to the retardance or DPPR, it is not possible to accurately determine the thickness of the blood vessel wall (as shown in Fig. 5).
Furthermore, the DPPR/UD and the wall thickness measurements by themselves did not provide as much contrast as the BBI for a hypertension disease biomarker. From the histograms in Fig. 1 and Fig. 2, threshold values were obtained using ROC curves and listed in Table 1. Although the results of both DPPR/UD and thickness measurements are statistically significant (p-values < 0.001), there is overlap between the data extracted from normotensive and hypertensive subjects. In addition, pdf-CI calculated for thickness and DPPR/UD (middle and bottom panel of Fig. 4) do not provide enough confidence as they show larger overlaps when compared to those of BBI (upper panel of Fig. 4). In arteries, the overlap in pdf-CI values are: BBI: 0.7% compared to DPPR/UD: 35.6% and thickness: 5.8%. For veins, those values are: BBI: 5.1%, in comparison to pdf-CI values of DPPR/UD: 42.6% and thickness: 8.2%. Thus, performing a diagnosis based on just the DPPR/UD (= birefringence) threshold values or just the thickness threshold values will cause errors. In other words, discriminating healthy from hypertensive subjects is only feasible when a combination of both thickness and DPPR/UD is used. BBI measurements (an index combining thickness and DPPR/UD), on the other hand, provided a clear distinction between the two groups of subjects, since there was very little overlap between the histograms (Fig. 2B and 2C). Moreover, the performance of each method examined by ROC curves (Supplementary Fig. S1) also favors the BBI measurements with a sensitivity of 99% and a specificity of 100%.
Furthermore, the histograms in Fig. 2B and 2C suggest a narrow normal distribution for normotensive data, with BBIs ranging between 10 to ~ 60 m for artery walls and 20 to ~ 70 m for vein walls. Hypertensive BBI data, on the other hand, are widely distributed (also normally distributed). The recruited patients were likely in different stages of disease, and some had controlled hypertension, while the hypertension of others was not controlled. These observations are confirmed by the box plots of Fig. 2A.
One weakness of the current study may be the fact that the data analysis was not performed blindly. In our previous study we showed that the analysis was independent of the examiner who analyzed the blood vessels and extracted the data 20. In the current study, two examiners also performed the analysis and a combination of extracted data by each of these examiners is reported in the paper. In addition, the criteria for determining the edge of the blood vessels are defined as such to minimize any examiner-attributed errors (with two out of three criteria to be satisfied). Also, since there is a visible margin between the normo- and hypertensive results, not masking the results can only affect the thickness measurements so much, for instance by 2 or 3 µm, which has only a small effect on the thickness, birefringence and BBI results. This would be a bigger issue if the two distributions were closer together. More importantly, the DPPR/UD of the blood vessel walls is determined automatically with a least-squares fit and is robust over a wide range of chosen vessel wall thickness.
Another perceived weakness may be the limited number of subject eyes that were imaged in this study (hypertensive, N = 11; healthy subjects N = 11). Note that for each eye, 4 blood vessels were analyzed and that often both eyes were imaged. Optical properties of the investigated blood vessels between hypertensive and healthy subjects were statistically well separated to limit the number of patients to eleven subjects in each category. The number of subjects recruited in our study closely matches those from one of the most recent glaucoma and diabetic retinopathy clinical studies 39. Indeed, there are some studies performed with PS-OCT that include more subjects, but these studies only performed experiments on patient eyes 40,41.
Another perceived weakness is a lack of characterization of the subjects. The lack of detailed characterisation of the hypertensive subjects such as nonavailability of a 24-hour ambulatory monitor blood pressure record is a limitation but all subjects with hypertension had an established physician diagnosed long-term hypertension and were on medication(s) to normalize the blood pressure. Furthermore, we did not measure the blood pressure of the healthy subjects. If these subjects were hypertensive and were not aware of it, their data would have moved towards the hypertensive distribution, decreasing sensitivity (99%) and specificity (100%). They could not have increased them much further. Similarly, we did not fully characterize the hypertensive patients, for instance by measuring their blood pressure over a 24-hour period, but if they were not hypertensive (and claimed to be hypertensive), their measurements would have led to a reduction in sensitivity and specificity.
While the current study presents the technology and the significance of the PS-OCT technology for the quantification of hypertension in small groups of healthy subjects and patients with hypertension, future studies will include more subjects with a wide spectrum of well-characterized blood pressure abnormalities, ranging from newly diagnosed untreated hypertension patients to patients with a long established history of treated and untreated hypertension for non-invasive diagnosis of pathogenesis caused by high blood pressure.