Stratification of Microvascular Disease Severity in the Foot Using Spatial Frequency Domain Imaging

Abstract

Background:

Microvascular disease (MVD) describes systemic changes in the small vessels (~100 um diameter) that impair tissue oxygenation and perfusion. MVD is a common but poorly monitored complication of diabetes. Recent studies have demonstrated that MVD: (i) is an independent risk factor for ulceration and amputation and (ii) increases risk of adverse limb outcomes synergistically with PAD. Despite the clinical relevance of MVD, microvascular evaluation is not standard in a vascular assessment.

Methods:

We evaluated 299 limbs from 153 patients seen clinically for possible lower extremity PAD. The patients were assessed by ankle brachial index (ABI), toe brachial index (TBI), and spatial frequency domain imaging (SFDI). These measurements were evaluated and compared to patient MVD status, defined by clinical diagnoses of (in ascending order of severity) no diabetes; diabetes; diabetes + neuropathy; diabetes + neuropathy + retinopathy.

Results:

SFDI-derived parameters HbT1 and StO2 were significantly different across the MVD groups (P < .001). A logistic regression model based on HbT1 and StO2 differentiated limbs with severe MVD (diabetes+neuropathy+retinopathy) from the larger group of limbs from patients with only diabetes (P = .001, area under the curve = 0.844). Neither ABI nor TBI significantly differentiated these populations.

Conclusions:

Standard assessment of PAD using ABI and TBI are inadequate for detecting MVD in at-risk populations. SFDI-defined HbT1 and StO2 are promising tools for evaluating MVD. Prospective studies with wound-based outcomes would be useful to further evaluate the role MVD assessment could play in routine clinical evaluation of patients at risk for lower extremity complications.

Introduction

Microvascular disease (MVD), also known as small-vessel disease or microangiopathy, describes a range of conditions which impact the arterioles, venules, and/or capillaries—vessels with diameters on the order of 100 um. These small-vessel disorders are systemic conditions with symptoms that manifest locally in organ-specific diseases. Local diseases related to MVD include retinopathy in the eye, nephropathy in the kidneys, coronary microangiopathy in the heart, neuropathy in the peripheral nervous system, and others.^1-6 In clinical practice, MVD is most common in patients with diabetes, where uncontrolled blood glucose levels cause chronic cellular stress. ⁷ In retinopathy, these accumulated stresses impair function of the neurons and

vascular cells, eventually causing blindness. ⁸ In the dermis, the response to sustained cellular stress includes an increase in thickness of the capillaries concomitant with a decrease in capillary lumen diameter and reduced perfusion capacity. ⁹ Impacts of the increased cellular stresses are not limited to the vascular cells; in diabetic polyneuropathy, both sensory and autonomic nerve function can be impaired as well. ¹⁰ To study MVD across large patient populations, existing diagnoses of diabetes, neuropathy, nephropathy, and/or retinopathy are commonly employed as indicators of systemic MVD.^1,11,12

The impact of small vessel dysfunction on health outcomes is increasingly appreciated. A study by Beckman et al ¹³ reported that the presence of clinically-defined MVD increased peripheral amputation risk independently of PAD (3.7 times higher than no MVD) and worked synergistically with PAD (amputation risk 13.9 times higher than baseline) to yield a 22.7 times higher amputation risk in patients that had both MVD and PAD. MVD has also been shown to correlate with increased rates of diabetic foot ulcer (DFU) development. ¹⁴

Despite the significant health risks associated with MVD, microvascular evaluation is not standard in a peripheral vascular assessment, often performed with ABI and/or TBI.^15-17 In this observational study, we assess the outputs of a vascular assessment within a population of patients with and without diabetes who were referred to the vascular clinic for assessment of PAD. The vascular assessment included ABI, TBI, and oxygenation and perfusion values measured using an imaging technology called spatial frequency domain imaging (SFDI).^18,19 The vascular assessment outputs were then compared to the patient MVD status, as defined by correlated clinical diagnoses, to analyze the efficacy of each technique for monitoring changes in the microvasculature.

Methods

All data acquired as a part of this study was approved by the Institutional Review Board of the University of Arizona.

Patients

From June 2016 to April 2020, we enrolled 154 total patients in this prospective, observational study. All enrolled patients consented to the study and had standard-of-care non-invasive vascular tests (waveform analysis and measurement of ABI, TBI in each limb—see Non-invasive Vascular Testing section below) and SFDI performed on their lower limbs at the Banner-University of Arizona Hospital. The patients’ previous diagnoses were noted at time of enrollment in the study and confirmed in their electronic health record. Patients were considered diabetic if they were taking medications for diabetes treatment. Patients were not offered study enrollment if ABIs and TBIs could not be obtained in either limb, or if both limbs were amputated. The breakdown by patient demographic information is shown in Table 1.

Table 1.

Information About the Demographics and Previous Diagnoses of the Included Patients.

Patient information and existing diagnoses
Total patients	154
Age in years (mean ± std. dev.)	68.1 ± 11.1
Men (Np, %)	80 (51.9)
Diabetes (Np, %)	56 (36.4)
Neuropathy (Np, %)	34 (22.1)
Retinopathy (Np, %)	14 (9.1)
Chronic kidney disease 5 (CKD5) (Np, %)	18 (11.7)

N_p: number of patients.

Considering that some patients (N_p = 9) had serious amputations (more proximal than a transmetatarsal amputation), vascular test data was captured in a total of 299 unique limbs (N_L: 299 limbs).

Microvascular Disease Spectrum

In this paper, we used the existing clinical diagnoses of diabetes, neuropathy, and retinopathy to represent MVD severity. These diagnoses were naturally conducive to a spectrum, as all patients with neuropathy had diabetes, and all patients with retinopathy had neuropathy (and diabetes). However, CKD5 diagnoses occurred independently of the other diagnoses, which suggested that kidney disease/nephropathy may reflect an alternate disease process. Due to this ambiguity, we did not include patients with CKD5 in the analysis, and in this paper analyzed the ABI, TBI, and SFDI outputs against the below MVD spectrum:

1. No Diabetes

2. Diabetes

3. Diabetes + Neuropathy

4. Diabetes + Neuropathy + Retinopathy

For subsequent analysis, we used the final group – diabetes + neuropathy + retinopathy – to indicate the highest-risk patients, those with “Severe MVD.”

Data Collection

Non-invasive Vascular Testing

Patients had brachial pressures measured in both arms, unless it was contraindicated, and the higher brachial pressure was used for the ABI calculation. Ankle pressures were measured in the dorsalis pedis and the posterior tibial arteries; the higher arterial pressure was used in ABI calculation. First-toe pressures were measured by plethysmography after ABI, where possible, to compute a TBI. All testing was performed in an accredited vascular lab; see Weinkauf et al ¹⁸ for more information.

Spatial Frequency Domain Imaging (SFDI) Technology

A research SFDI system (Reflect RS; Modulim: Irvine, CA) was used to image the patients in this study. Briefly, the SFDI system used a combination of flood and structured illumination at multiple wavelengths in the visible and near-infrared light spectrum (λ = 450-900 nm) to quantify skin tissue reflectance (Figure 1a and b). These projection-and-capture sequences took <10 seconds to complete, after which the reflectance data was processed to yield the oxygenation and perfusion maps (eg, Figure 1c). Because of the smaller field of view of the research system used in this study, 2 images were required to capture the full foot; the 2 images were stitched together to yield a full-foot image prior to analysis.

Figure 1.

An Illustration of the oxygenation and perfusion maps extracted by SFDI in an example plantar foot. (a) Depicts the sampling of tissue reflectance at visible and near infrared wavelengths, combined with the (b) structured illumination of SFDI. (c) Shows example color-coded maps of the tissue oxygen saturation (StO2) and the papillary/superficial hemoglobin counts (HbT1) from the representative volume of skin tissue.

We have previously demonstrated the measurement of hemoglobin oxygenation and distribution markers from SFDI method. ¹⁸ This manuscript focuses on 2 biomarker/chromophore outputs from SFDI that have previously been found to be statistically different between diabetes and non-diabetes groups:^18,20

• - HbT₁: The papillary hemoglobin; the amount of hemoglobin in the capillary-rich superficial layer of the dermis (1-2 mm deep)

• - StO₂: The tissue oxygen saturation (%); the percent of oxygen-bound hemoglobin in the full-thickness of the dermis (1-4 mm deep).

StO2 measurements differ from oxygenation measurements reported by TcpO2 or TCOM technologies. In TcpO2, pO2 measurements represent the local free oxygen present in very superficial tissue that diffuses through the skin to the contact probe. In SFDI, the StO2 represents the percentage of hemoglobin molecules in the mixed arterial/venous bed of the superficial and subsurface tissue that has oxygen attached. Because of these key differences, pO2 measurements and StO2 measurements are related but not necessarily correlated in disease states and population.

Vibratory Sensation Testing

We assessed sensory neuropathy (ie, loss of protective sensation) using a tuning fork vibration test on 3 anatomical locations—medial malleolus, hallux, and thumb. Briefly, we used 128 MHz tuning fork and timed the duration of sensation (or none) on all anatomic locations on the right and left.

Data Analysis

Image Analysis

For each foot and each chromophore (StO2 and HbT1), the SFDI output images were down-sampled into a single number by taking the median of all the tissue pixels.^18,20,21 This simple image analysis was preferred over more complex analyses in this relatively limited number of images (299 total limbs).

Tests of Statistical Significance

For statistical testing across MVD and sensation groups, the Kruskall-Wallis test was used, with a P-value less than .05 used to reject the null hypothesis of no differences across groups. If a P-value less than .05 was achieved, the Dunn test was used to identify the specific differences across groups.

Logistic Regression and Receiver Operator Characteristic (ROC) Curve Analysis

Logistic regression models were used to assess the value of the different metrics (ABI, TBI, HbT1, and StO2) for stratification of severe MVD in diabetic limbs. Various thresholds with corresponding sensitivities and specificities were explored graphically by receiver operator characteristic (ROC) curve analysis. In ROC curve analysis, the area under the curve (AUC) represents the overall predictive power of the measured variable. The AUC ranges from 0.5 (no predictive power) to 1.0 (perfect correlation). No test/train split or resampling was performed, due to the smaller sample size in the analysis. All analysis was performed using Python 3.7.

Results

Stratification of MVD Groups

The relationship between the existing MVD-related clinical diagnoses and the ABI and TBI for each limb are shown in Figure 2. There were no significant differences between MVD groups (P = .944, .946 for ABI, TBI respectively).

Figure 2.

Stratification of MVD-related clinical diagnosis groups by standard-of-care. (a) ABI and (b) TBI.

n denotes number of limbs.

In contrast to the ABI and TBI, the SFDI-derived papillary hemoglobin (HbT1) decreased concomitantly with an increase in the tissue oxygen saturation (StO2) as MVD severity increased (Figure 3). The differences between groups were statistically significant for each SFDI metric (P < 10^-3 for both HbT1, StO2). A pictorial representation of the correlation between HbT1 and the defined MVD spectrum is shown by the color-coded HbT1 maps in Figure 3a. The Dunn test revealed that the differences between all groups were statistically significant for both HbT1 (Figure 3c) and StO2 (Figure 3d), except for the difference between the diabetes group and the diabetes + neuropathy group.

Figure 3.

Stratification of MVD-related clinical diagnosis groups by SFDI-derived (a) HbT1 and (b) StO2. (c) and (d) show the Dunn Post-Hoc test to isolate significant differences across groups.

n, number of limbs.

A pictorial representation of the correlation between HbT1 and the defined MVD spectrum is shown in Figure 4. Example limbs were selected to best represent the mean HbT1 value of each MVD group, as shown in Figure 3a. Standard RGB color images (Figure 4a) demonstrate no clear changes as the MVD severity increases, whereas the HbT1 maps (Figure 4b) show a gradual decrease in papillary hemoglobin.

Figure 4.

Representative color and HbT1 images from each MVD-related clinical diagnosis groups. (a) shows the 4 representative color images, while (b) includes the associated HbT1 images, which illustrate a decrease in median HbT1 as MVD-related diagnoses increase. For future analyses, the clinical diagnosis groups are sorted into Other Diabetes and Diabetes with Severe MVD, as shown below (b).

Prediction of Severe MVD

In the subsequent analysis, we divided the diabetic limbs into 2 groups—Other Diabetes and Diabetes with Severe MVD – using retinopathy to indicate severe MVD.

ROC analysis was performed using TBI, and then a combination of TBI & ABI (Figure 5a). The low AUC values (0.525 for TBI, 0.605 for TBI & ABI) indicate poor predictive power of ABI and TBI for identifying limbs with severe MVD (Figure 5a and b).

Figure 5.

Binary classification for separating the Diabetes with Severe MVD from the Other Diabetes limbs. (a) depicts the receiver operating characteristic (ROC) curves for logistic regression models based on TBI and a combination of TBI & ABI. (b) shows the relationship between ABI and TBI for both groups of limbs. Similarly, (c) depicts the ROC curves for models based on HbT₁ and a combination of HbT₁ & StO₂, with hollow circles indicating the “best threshold” based on equal-weighted sensitivity and specificity (AUC: Are Under the Curve). (d) shows the decision boundary for the prediction model 1 (based only on HbT1) and model 2 (based on HbT₁ & StO₂), as well as a theoretical clinical model based on consecutive thresholds by HbT₁ then StO₂.

A similar analysis was performed using the SFDI-derived metrics (Figure 5c and d). The SFDI-derived papillary hemoglobin (HbT1) was used to develop a simple logistic regression predictive model, followed by a combination of HbT1 & StO2 (Figure 5c). The HbT1 model achieved an AUC of 0.812, whereas the model with HbT1 & StO2 achieved an AUC of 0.844. For demonstrative purposes, the circles labelled 1 and 2 in Figure 5c represent the “best prediction thresholds,” considering an equal desire for sensitivity and specificity. The scatterplot and prediction boundaries in Figure 5d highlight the consistently low HbT1 and high StO2 in limbs with severe MVD.

Correlations with Sensory Neuropathy

To assess loss of vibratory sensation, which is associated with sensory neuropathy, each patient’s limb sensation was measured by tuning fork timing analysis at the right and left hallux, malleolus, and the interphalangeal joint of the thumb (see Vibratory Sensation Testing section above).

The vibratory sense times showed no clear correlations with ABI, TBI, or SFDI-derived HbT1 and StO2. However, HbT1 was significantly different in those with a non-sensate hallux, or a combination of a non-sensate hallux and a non-sensate malleolus (all thumb were sensate) compared to those with sensation (P = .010, Figure 6c and d). The ABI, TBI, HbT1, and StO2 within these 3 groups of diabetic limbs—sensate, non-sensate hallux, and non-sensate hallux and malleolus – are explored graphically in Figure 6.

Figure 6.

Assessment of sensation loss in diabetic limbs by ABI, TBI, and HbT1. ABI (a) and TBI (b) show no statistical differences across sensation groups (P > .05), whereas HbT1 (c) finds statistical differences across the sensation groups. The Dunn test (d) reveals that HbT1 is significantly lower (P = .010) in diabetic limbs with no sensation in malleolus and hallux than diabetic limbs with sensation in both regions.

*Note that there are fewer data points in plots assessing ABI and TBI than in plots assessing SFDI outputs, due to limbs from which TBI or ABI could not be obtained.

Discussion

Key Findings

Stratification of MVD Groups

Our data suggest that standard-of-care vascular tests (ABI and TBI) do not correlate with known clinical diagnoses that support the presence and severity of MVD (Figure 2). These data agree with the notion that ABI and TBI are generally used for evaluation of large-vessel arterial flow, rather than diseases of the small-vessels. ²² In practice, most physicians do not directly measure perfusion in the small vessels. As such, it is difficult to consider the patient’s microvascular status in their treatment plan and estimation of patient limb risk. Our data indicate that SFDI-derived HbT1 and StO2 correlate with the patient’s MVD-related diagnoses (Figures 3-5), and that these metrics could isolate patients with the most severe MVD. A clinical parameter to represent and monitor MVD is highly relevant to lower limb health, as it was recently shown that MVD increases risk of amputation independently from PAD and has synergistic effects with PAD resulting in a 22-fold increased risk of amputation. ¹²

Isolation of Severe MVD

It has been shown that patients with multiple MVD-related diagnoses are at higher risk of ulceration. ¹⁴ Therefore, it is most relevant to identify and elevate care for diabetes patients with multiple MVD-related diagnoses, such as the patients with Severe MVD (diabetes, neuropathy, and retinopathy) assessed in Figure 5. Models based on (i) HbT1 and (ii) a combination of HbT1 & StO2 isolated limbs with severe MVD (AUC of 0.812 and 0.844, respectively; Figure 5c). A simple application of this result is suggested by the Theoretical Clinical Model line in Figure 5d, which shows how a model based on thresholding median HbT1 and median StO2 could isolate patients with severe MVD for elevating care.

Correlations with Sensory Neuropathy

Our analysis of limb sensation via neuropathy timings indicates that HbT1 shows promise for assessing microvascular changes associated with loss of sensation in diabetic limbs (Figure 6c). Loss of sensation in sensory neuropathy is a late-stage diagnosis; as described by Malik, ²³ it is preceded by changes in the small nerve fibers, at a stage when changes may be reversible.^23-26 Future work involving a better assessment of nerve fiber dysfunction would be useful to further evaluate this aspect of disease severity and indicate if such changes in small fibers can be tracked via changes in the microvascular perfusion and oxygenation.

Our work did not find any parameters that correlated with the timed duration of sensation. This may be because the sensation timing was dependent upon patient cooperation and the consistency of measurement/placement.

Pathophysiology of MVD

The main finding of this study—that MVD correlates with reduced hemoglobin in the papillary dermis (lower HbT1) and higher tissue oxygen saturation (higher StO2)—sheds light on the intrinsic pathological processes in microangiopathy. These observations suggest an MVD-driven impairment in the tissue’s ability to metabolize oxygen. However, the micro-structural alterations that drive these changes in tissue oxygenation and perfusion are not immediately evident from SFDI. Other studies have shown that MVD can thicken the arterioles and capillaries, in a process akin to atherosclerosis in the larger vessels. This change is known as capillary basement membrane (CBM) thickening. ²⁷ CBM thickening reduces the capillary lumen diameter, impairing healthy vasodilation/vasoconstriction processes in the cutaneous microcirculation. The reduced lumen diameter of these small vessels may be a contributor to the reduced hemoglobin in the papillary dermis (HbT1). Other studies have demonstrated the presence of arteriovenous (AV) shunts in people with diabetes with neuropathy, diverting oxygenated blood away from the capillaries and back into the venous return system. ²⁸ Known as “AV Shunting” or “Capillary Steal,” this bypass results from dysfunction in the nerves controlling the arteriovenous sphincter. These 2 phenomena—CBM thickening and AV shunting—are possible contributors to the impaired tissue oxygen offloading in MVD. The complete microvascular pathology in diabetes/diabetic polyneuropathy is manifold and outside this scope.^9,29 However, the presence of hemoglobin in the capillaries, as indicated by HbT1, serves as an important biological marker, aggregating the effects of microstructural changes into a combined measure of MVD severity.

Study Limitations and Extensions

A limitation of this study was the lack of direct measurement of microvascular changes in the dermis via invasive biopsy (ie, capillary lumen diameter, capillary density). Biopsies present challenges, and there are risks of biopsy-induced wound healing failure. However, a thorough understanding of the structural changes accompanying the progression of MVD is essential for advising therapies. ³⁰ A possible design could include assessing patients with existing wounds to compare to quantitative microvascular measurements.

Like existing techniques such as ABI and TBI, SFDI is a static quantification of the microvascular state. As comparable studies have examined how ABI changes during exercise, future studies could examine how SFDI metrics change in response to local heating or stimulation. Quantification of microvascular responsivity via repeated SFDI measurement may highlight a patient’s response to wound development or dermal mechanical fatigue.

Future studies would benefit from a direct assessment of sensory neuropathy, autonomic neuropathy, retinal exams, and A1C history to categorize subjects’ disease severity. The current design focused on chart review due to limitations in the recruitment time. Although SFDI showed promising potential to quantify MVD in this population of patients additional studies evaluating wound healing and/or ulceration outcomes would be very useful clinically. Currently, the microvascular status of a patient with diabetes is a physician-dependent judgment. Tools like SFDI may provide needed quantification and standardization in assessment of these high-risk populations.

Conclusions

For a more complete understanding of limb health and risk of ulceration/amputation, the limb’s microvascular health must be considered. In fact, the microvascular health is intimately tied to the tissue health—without oxygen offloading in the capillaries, the tissue fails to function. ¹² This is especially evident in the dermis of the lower limb, where microcirculatory compromise can impair the tissue maintenance and wound healing processes. To appropriately assess the tissue health, it is essential to examine not only the upstream flow, but also the microvascular oxygenation and perfusion. This study has shown how ABI and TBI are not suited to this task. Conversely, SFDI assessment of the plantar foot demonstrated the ability to stratify different MVD groups. Microvascular assessment using tools like SFDI may help clinicians to selectively identify and elevate care for patients at the highest risk.