Non-invasive Diagnosis and Prognosis Values of 3D Pseudocontinuous Arterial Spin Labeling and Optical Coherence Tomography Angiography in Proliferative Diabetic Retinopathy

Background: 3D Pseudocontinuous Arterial Spin Labeling (3D-PCASL) MRI and optical coherence tomography angiography (OCTA) have been applied to detect ocular blood flow (BF). We aim to characterize the ocular BF in diabetic retinopathy (DR) using 3D-PCASL and OCTA, to discuss the relationship between ocular and cerebral BF, and to evaluate their potential utility to assess the severity of DR. Methods: A total of 66 participants (132 eyes) were included. Seventy-two eyes were classified in the proliferative diabetic retinopathy (PDR) group, and 60 were in the non-proliferative diabetic retinopathy NPDR group. Ocular and cerebral BF values were detected by 3D-PCASL using a 3.0T MRI scanner with two post-labeling delays (PLDs). Vessel density (VD)/perfusion density (PD) of the macular or peripapillary area were detected by OCTA. Parameters and clinical characteristics were compared between the PDR and NPDR eyes utilizing two-sample t-tests and chi-square tests. Spearman's rank correlation analysis, logistic regression analysis, and receiver operating characteristic curves (ROC) analyses were performed to evaluate the factors' role in DR severity. Results: The perfusions of the retinal/choroidal plexus (RCP), optic nerve head (ONH)/optic nerve (ON), and VD/PD of macular/peripapillary area in the PDR group were significantly lower compared to the NPDR group (p < 0.05). They were protective factors for PDR [ORs = 0.842 for RCP (1.5 s PLD), 0.910 for ONH (1.5 s PLD), 0.905 for ON (both 1.5 and 2.5 s PLD), 0.707 for macular VD, 0.652 for peripapillary VD, p < 0.05, respectively]. Ocular BF had a positive correlation with BF of the occipital lobe (OL) and temporal lobe (TL) in the cerebrum. The BF of RCP (lower than 7.825 mL/min/100 g at 1.5 s PLD) indicated PDR [areas under the curve (AUCs) = 0.682, 95% CI: 0.588–0.777, sensitivity: 70.7% specificity: 63.9%]. The AUC of RCP (PLD = 1.5 s) BF combined with peripapillary VD was 0.841 (95% CI: 0.588–0.777, sensitivity: 75.9% specificity: 82.9%). Conclusions: 3D-pcASL and OCTA may be effective non-invasive methods to measure ocular blood flow in DR patients and assess the severity of DR.

Conclusions: 3D-pcASL and OCTA may be effective non-invasive methods to measure ocular blood flow in DR patients and assess the severity of DR.

INTRODUCTION
Diabetic retinopathy (DR) is an ocular manifestation that affects one-third of diabetes patients (1). It is also the leading cause of visual impairment and new-onset blindness in working-age adults (2). DR is characterized by microvascular dysfunction (3,4), and, if left untreated, can result in retinal neovascularization, loss of vision, and pain. More than 90% of blindness cases caused by DR, however, can be prevented by early detection and appropriate intervention (5).
At present, fluorescein angiography (FA) is the gold standard method used to evaluate the severity of DR-but, due to its invasive nature, depth limitations, ocular media restrictions, and the adverse effects of dyes (6), it is unsuitable for either early diagnosis of DR or long-term monitoring of the condition. Recent work suggests that imaging methods focusing on neural blood perfusion may have predictive value in evaluating DR. Prior evidence shows that ocular vascular changes [i.e., hypoperfusion of the retina /choroid plexus (RCP) (7)], rather than structural lesions, may underly the progression of DR (8)(9)(10)(11)(12)(13). Retinal blood perfusion is controlled by neurons. In one study in a DR model, the ocular arteriole and capillary failed to dilate in coordination with metabolic demands when neurons were stimulated by lights (14). This lack of coordination could lead to a cycle of neuronal dysfunction and neurodegeneration, leading to vision loss and an increasingly pro-inflammatory environment which could ultimately promote the progression of DR (15). Thus, doppler, laser speckle, and angiography techniques have been used to measure ocular blood flow (16). However, quantifying images, particularly of the choroid and optic nerve, is still challenging using these methods.
Arterial spin labeling (ASL) perfusion MRI is a relatively new, versatile, and non-invasive imaging technique that uses magnetically labeled water in arterial blood as an endogenous tracer. Because of its highly layered structure, ASL-MRI is particularly applicable for measuring RCP perfusion (17). In a murine model of DR, Muir et al. (18) found that both retinal and choroid blood flow was successfully detected by ASL, with a larger view field and without depth limitation. Peng et al. (19) have also recently used the novel pCASL technique to acquire ocular blood flow images and parameters in humans with DR. Other ocular vasculature profiles in DR, such as vessel density (VD) and perfusion density (PD), can be detected by OCTA, an alternative non-invasive technique (20,21).
Here, we use 3D-pCASL and OCTA to quantify RCP and ON blood perfusion changes throughout the progression of DR, and evaluate their potential utility to assess DR severity. Diagnosis of type 2 diabetes mellitus (DM) in each patient was confirmed by an internal medicine consultant. DR was confirmed by an ophthalmologist using fundus imaging and fluorescence fundus angiography (FFA). All recruited patients underwent detailed ophthalmological investigations, including best-corrected visual acuity (BCVA)-recorded as the logarithm of the minimum angle of resolution (logMAR)-slit-lamp biomicroscopy, intraocular pressure measurements, and fundus color photography (Kowa Nonmyd WX; Kowa Company Ltd., Japan) after full mydriasis and FFA. FFA images were read by a retina-trained specialist.

MRI Data Acquisition
MR imaging was carried out using a 3.0 T Ingenia scanner (Philips Healthcare, Best, Netherlands) equipped with a commercial body coil for transmission and a 16-channel head coil for reception. Conventional T2-weighted fast-spinecho and 3D time-of-flight (TOF) MR angiography images were obtained before the ASL sequence to exclude patients with brain lesions or arterial stenosis. All patients were required to keep their eyes closed and stay motionless during the examinations.
Blood flow (BF) was imaged utilizing the pcASL technique. The gradients adopted an unbalanced scheme and the labeling

ASL Data Quantification
The RCP blood flow map was automatically derived from the ASL images using a dedicated workstation (IntelliSpace Portal Release v. 7.0.4.20175, Philips). A region of interest (ROI) in the RCP map was placed independently by an experienced radiologist and an experienced ophthalmologist at the level of the optic nerve. The two specialists reached an agreement on how to draw ROIs, which were two-dimensional and covered as much of the RCP as possible. The average value of each ROI was calculated automatically and then analyzed.

Optical Coherence Tomography Angiography (OCTA) Measurement
OCTA examination was completed using the 68 kHz Zeiss Cirrus HD-OCT 5000 with AngioPlex at a wavelength of 840 nm (Carl Zeiss Meditec, Dublin, CA, USA) (22). Volumetric scans were processed using optical microangiography (OMAG) algorithms to generate flow images. The OMAG algorithm analyzes differences in both phase and intensity information from repeated B-scans to quantify motion contrast. The superficial capillary plexus (SCP) was extended from the internal limiting membrane to the inner plexiform layer.
A macular and optic disc-centered 6 × 6 mm scan was taken to investigate microvasculature. Each 6 × 6 mm scan contained 350 A-scans in each B-scan, and each B-scan was repeated two times at the same location. The vessel density (VD) was defined as the total length of perfused vasculature per unit area in a region of measurement, and perfusion density (PD) was defined as the total area of perfused vasculature per unit area in a region of measurement. Cirrus OCTA software (ver. 10.0) automatically calculated the value of the SCP according to the Early Treatment of Diabetic Retinopathy Study (ETDRS) subfields. We analyzed the macular VD, macular PD, peripapillary VD, and peripapillary PD in SCP of the full field of view (FOV).

Statistical Analysis
All statistical analyses were performed using SPSS statistical software (version 26.0, SPSS Inc., Chicago, Illinois, USA). Data were expressed as the mean ± standard deviation. Twosample t-tests and chi-square tests were performed to compare differences between the NPDR and PDR groups. Spearman's correlation analyses were utilized to evaluate the correlation between BF in the RCP and cerebrum. Binary Logistic regression analysis was used to obtain odds ratios (ORs). Diagnostic accuracy of RCP blood flow in DR grades that were measured by 3D-pCASL and OCTA was evaluated using receiver operating characteristic (ROC) curves. P-values < 0.05 were considered statistically significant.

Patient Demographic and Clinical Characteristics
A total of 66 DR patients (132 eyes) were included, with the PDR group consisting of 38 patients (72 eyes), and the NPDR group consisting of 32 patients (60 eyes). Four patients had one eye included in the PDR group and the other in the NPDR group. There were 45 eyes in group A and 22 eyes in group B. No significant differences were identified between two groups in BCVA, IOP, age, gender, or other clinical indexes ( Table 1).

Comparison BF and VD/PD Values of RCP and ON
The BF values of RCP and ONH in PDR group were significantly lower than the NPDR group at 1.5 s PLD, but not at 2.5 s PLD (RCP, 7.19 ± 3.30 vs. 9.93 ± 5.26 mL/min/100 g; ONH, 7.78 ± 4.38 vs. 10.23 ± 6.53 mL/min/100 g p < 0.05, respectively). The BF values of ON in the PDR group were notably lower than the NPDR group at 2 PLD (PLD = 1.5 s, 7.12 ± 3.18 vs. 9.24 ± 6.53 mL/min/100 g; PLD = 2.5 s, 10.43 ± 5.31 vs. 13.91 ± 6.86 mL/min/100 g p < 0.05, respectively). There were no differences between the BF values of OL and TL between the two groups at either PLDs (Figure 1)   The average VD and PD of the macular and peripapillary area in the PDR group were remarkably lower than in the NPDR group (P < 0.001 for both; see Table 2, Figure 2).

The Relationship Between BF of Ocular and Cerebrum
Spearman correlation analysis was performed to explore the relationship between ocular and cerebral perfusion ( Table 3). Spearman's correlation coefficients between BF of RCP/ONH/ON and TL at 1.5 s PLD were highly significant, with values of 0.580, 0.458, and 0.385 (P < 0.001 for all). Spearman's correlation coefficients between BF of RCP/ONH/ON and OL at 2.5 s PLD were also highly significant, with values of 0.239, 0.403, and 0.244 (P = 0.006, P < 0.001, P = 0.005, respectively).

Logistic Regression Analysis of Factors Associated With PDR
Hierarchical data logistic regression was used to examine the association between BF of RCP/ONH/ON and VD/PD of  Figure 3).

DISCUSSION
Here, we detected BF within the RCP, ON, and cerebrum in different stages of DR using multi-PLD 3D-pCASL, and measured the VD/PD of SCL in the macular and peripapillary areas using OCTA. As DR progresses, we found that BF of the RCP and ONH significantly decreases at 1.5 s PLD, and BF of the ON decreases at both 1.5 and 2.5 s PLDs. We also found that BF of the ON in PDR patients with severe visual impairment was much lower at   2.5 s PLD. Hypoperfusion in the ocular and ON were risk factors for the development of PDR. Furthermore, the decreased BF of RCP/ONH/ON, and the decreased VD/PD of SCP in both the macular and peripapillary areas indicated more advantage stage of DR. BF of the RCP had the highest diagnostic efficiency at 1.5 s PLD, followed by ON at 2.5 s PLD in PDR patients. The two noninvasive methods combined had higher AUC, sensitivity, and specificity in diagnosing PDR than either method alone. Interestingly, as DR progressed, both cerebral blood flow (CBF) and ocular blood flow decreased. The cerebrum may play a key role in vision formation and transmission. Previous studies (23,24) have demonstrated that vision impairment causes reductions in cerebral BF, as well as functional changes, especially in vision-related areas such as the TL and occipital lobe (OL). However, other studies (25,26) have found increased CBF in visually impaired patients. In our research, the RCP BF had a positive correlation with TL and OL BF at both PLDs. CBF had a decreasing trend with DR severity, but there was no statistical difference between PDR and NPDR groups. A possible explanation for this finding is that ocular adjustment and adaptation in DR can trigger a compensatory process that increases blood perfusion within the visual cortex.
One impact of DR is that normal retinal vascular autoregulation is disturbed. Vasoconstriction, provoked by protein kinase C or renin-angiotensin (27), ultimately leads to capillary occlusion and tissue hypoperfusion, or even nonperfusion. Neuronal dysfunction in DR may appear earlier than microvascular lesions, and DR may progress to advanced stages without producing any immediate symptoms. Thus, it is critical to determine appropriate intervention points, as well as to determine whether further invasive treatment methods would be able to prevent vision loss in DR patients. Current methods classify DR using fundus photography (28,29) on the basis of the severity of certain clinical hallmarks (i.e., microaneurysms, capillary non-perfusion, leakage, and neovascularization). However, the fluorescein dye used in this process may lead to nausea, vomiting, itching, urticaria, and even anaphylaxis (6), and is unable to be used in vulnerable patients populationsincluding the elderly, children, patients with difficult venous access, and patients with renal insufficiency. Furthermore, FA detects only ∼30% of the total retinal surface area (5). Other previous methods used to examine DR progression (30) include measuring ocular blood flow in different stages of DR by fundus photography, Doppler velocimetry, the blue field entoptic technique, and pneumotonometry. Using the Doppler device, Cuypers (8) and Nagaoka and Schocket et al. (7,12) found that RCP BF was lower in PDR than in NPDR patients.
Other research groups have used ASL perfusion MRI to estimate CBF (31). With ASL, BF can be measured quantitatively in units of mL/min/100 g. The ASL technique is superior to existing perfusion imaging techniques because it is generally more sensitive to smaller vessels. It has a larger field of view with no depth limitations (32)(33)(34), meaning that BF in the back of the ocular field can be fully evaluated. Using ASL blood-flow MRI, Eric R (18) found that, in a mouse model of diabetes, reduced choroid and retinal ocular BF was an indicator of DR. Peng and Zhang (19,34) were the first to investigate the feasibility of ASL to measure BF in the normal human retina. As a recently developed ASL technique, pCASL uses high sensitivity array coils, high SNR, and tagging efficiency to overcome the limitations of PASL and CASL. Further, 3D-pCASL is recommended in clinical studies (35,36).
OCTA is another alternative non-invasive method that has been widely used in DR patients. It utilizes the motion of erythrocytes to detect BF in the retinal capillaries and potentially provides insights into retinal pathophysiology. Several researchers have detected microvascular details and capillary changes in DR patients (20,21,(37)(38)(39)(40). Hwang et al. (21) found VD and perifoveal PD were significantly reduced in the more advanced stages of DR (41,42). Our study showed that VD and PD of both macular and peripapillary areas in PDR patients were reduced compared to NPDR patients.
Various forms of diabetic optic neuropathy can also accompany the progression of DR, and can eventually threaten vision (43). The optic nerve is impacted by impaired ocular microcirculation in DR. ONH perfusion can be measured by OCTA, but few techniques are available for the optic nerve because of its deep location. Here, we successfully quantified ON BF using a 3D-pCASL technique that has a large field of view without depth limitations. Perfusion of the ON was lower in PDR than NPDR eyes, and even worse in PDR eyes with low visual acuity.

CONCLUSION
Our findings suggest that multi-PLD 3D-pcASL and OCTA offer non-invasive, replicable, and highly diagnostically efficient methods for tracking perfusion changes in the RCP and ON. They may also be reliable techniques to assist in categorizing DR stage and monitoring disease progression.

LIMITATIONS
Our study utilized a retrospective design and was limited to a single center and absence of control group. Stringent inclusion criteria also limited our sample size to 132 eyes, which may have underpowered some aspects of the study. Due to the retrospective nature of the study, we could not achieve uniform ASL acquisition at a set time point. There could also have been a delay between symptoms and inclusion in the study. Additionally, an admission rate bias for patients in whom ASL was performed may have prevented us from having patients with comparable illness severity. The possibility of a measurement bias for imaging is also possible whenever manual measurements are used.

DATA AVAILABILITY STATEMENT
The original contributions presented in the study are included in the article/supplementary material, further inquiries can be directed to the corresponding authors.

ETHICS STATEMENT
The studies involving human participants were reviewed and approved by This study was approved by the BFH Ethics Committee and was conducted following the latest iteration of the Declaration of Helsinki (version: 2019-P2-201-01). The patients/participants provided their written informed consent to participate in this study.

AUTHOR CONTRIBUTIONS
HL and JL was involved in the conception and design of the study. HW and ZW reviewed the orbit MRI images. YH and SS contributed to the acquisition, analysis and interpretation of data as well as drafting the manuscript and revising it critically. JS and YW were the duty of diagnosis the disease. All authors have provided final approval of the version to be published. All authors have agreed to be accountable for all aspects of the work in ensuring that questions related to the accuracy or integrity of any part of the work are appropriately investigated and resolved.

ACKNOWLEDGMENTS
Thank you MedEditing (www.medediting.com) for the English language editing service.