Exploring the Dynamics of Ischemia and Reactive Hyperemia With Skeletal Muscle Blood Oxygen Level Dependent MRI in Patients With Peripheral Artery Disease, Age‐Matched Controls, and Young Healthy Subjects

ABSTRACT

Background

Skeletal muscle blood oxygen level dependent (BOLD) MRI is a technique for assessing vascular function in peripheral limbs. In patients, however, an increased frequency of atypical response patterns has been observed, warranting investigation into its underlying causes.

Purpose

To explore the dynamics of cuff‐induced ischemia and reactive hyperemia with BOLD MRI, focusing on the occurrence, quantification, and interpretation of frequent atypical cuff‐induced response patterns.

Study Type

Secondary assessment of prospectively collected datasets.

Population

Seventeen peripheral artery disease (PAD) patients (age: 64–80 years, ankle‐brachial index (ABI) range: 0.4–0.8), 8 age‐matched controls (age: 64–82 years, ABI range: 1–1.2), and 14 young controls (age: 22–39 years).

Field Strength/Sequence

$T_2^{*}$‐weighted multi‐echo gradient‐echo sequence with 11 equidistant echo‐times at 3T.

Assessment

A transverse slice of the calf was imaged repeatedly during an upper‐thigh cuff compression cycle consisting of 1‐min baseline, 5‐min arterial occlusion (cuff inflated) and 5‐min reactive hyperemia (cuff deflated). $T_2^{*}$‐mapping with ROI‐wise analysis of the soleus and gastrocnemius muscles produced ${\mathrm{T}}_2^{*}$‐time curves and previously established metrics, including the hyperemic upslope (${\mathrm{HS}}_{\mathrm{up}}$) and time‐to‐peak (TTP) were assessed. The time‐curves were surveyed to identify the frequency and type of deviations from expected. ${\mathrm{T}}_2^{*}$‐time curves of soleus were further analyzed by fitting a parameterized function to derive additional metrics including depth of the undershoot on cuff release and deviation from a monotonic T2* decrease. Ankle‐brachial index (ABI) served as a clinical reference for comparisons.

Statistical Tests

Non‐parametric 2‐tailed Wilcoxon rank‐sum tests to assess differences between patients and age‐matched controls. A p value < 0.05 was considered significant.

Results

Atypical cuff‐induced response patterns in PAD patients included a negative cuff‐induced transient (42%, soleus) and non‐monotonic declines in $T_2^{*}$ during cuffing (58%, soleus). While these contributed to increased variabilities in patients, $\text{there were significant differences in}{\mathrm{HS}}_{\mathrm{up}}$ (−0.63 ms/s) and TTP (62.8 s) between patients and age‐matched controls.

Data Conclusion

HS_up and TTP provided means to quantify calf muscle responses to cuffing. Specific cases of non‐monotonous $T_2^{*}$ decrease during cuffing suggest the detection of venous valve leakages in inter‐muscular veins. Insufficient cuff pressure produced curves with absent ischemic plateau and weak hyperemic responses, the recognition of which is important to prevent physiological misinterpretations of ${\mathrm{T}}_2^{*}-\text{time}$ curves and derived metrics.

Evidence Level

1.

Technical Efficacy

3.

Plain Language Summary

Peripheral artery disease limits blood flow to the legs, affecting muscle function. This study used an MRI technique sensitive to blood oxygen levels to examine calf muscle responses when blood flow was briefly blocked and restored. Data from 17 patients, 8 older healthy adults, and 14 young participants were analyzed visually and semi‐quantitatively. Patients often showed atypical recovery patterns, including cases of incomplete occlusion. The early recovery phase after reperfusion best distinguished patients, with peak response occurring about 60 s later than in controls. These findings suggest this MRI method could be a useful, non‐invasive tool for assessing muscle perfusion.

1. Introduction

Over the last decade, blood oxygen level dependent (BOLD) MRI has gained traction as a potential technique for evaluating vascular function in peripheral muscle. Perfusion‐related metrics derived from BOLD MRI have been shown to differentiate between healthy controls and patients with diseases involving the circulatory system, such as peripheral artery disease (PAD) [1, 2, 3, 4, 5], chronic limb‐threatening ischemia (CLTI) [6], type 2 diabetes [7], and systemic sclerosis [8]. The BOLD MRI technique is non‐invasive but is commonly paired with an experimental design that can provoke changes in blood oxygenation to improve the image contrast in peripheral muscles of interest. Several approaches have been proposed, including muscular activity prior to and during scanning [9, 10]. In contrast, the use of a pressure tourniquet during the MRI session to produce ischemia and a subsequent reactive hyperemic response upon cuff release, requires no need for patient‐performed maneuvers and enables simultaneous measurement of dynamic BOLD signal changes throughout the cuffing procedure [11]. Cuff‐induced ${\mathrm{T}}_2^{*}$‐mapping BOLD MRI may have potential for phenotyping disease progression states and treatment outcomes, based on how it accentuates differences in physiologic responses between healthy and diseased individuals [9, 12, 13].

Previous studies with cuff‐based BOLD MRI have shown correlations between metrics derived from the hyperemic response and the current bedside reference test for diagnosing PAD, the ankle‐brachial index (ABI) [1, 13, 14]. In simplified terms, ABI is the ratio between systolic ankle and brachial blood pressure, measured at rest in a feet‐elevated pose [15]. While ABI is a highly accessible measure, it is primarily sensitive to conduit artery blood flow effects and is known to have low accuracy and reproducibility [15, 16], hence additional diagnostic tools are needed. Studies focusing on the variability in the BOLD dynamics within and between subjects are currently lacking. Measurements have previously been presented as averages over several individuals [1, 2, 17], or by single representative measurements [8, 18]. Further, some studies only show and derive BOLD metrics from the time point of the cuff release and onward [1, 17, 19].

Thus, the aim of this study was to descriptively and quantitatively characterize frequently occurring but previously neglected traits in BOLD curves from lower limb striated muscle. A second aim was to explore how these, together with previously established BOLD curve metrics, can facilitate the interpretation of observed responses to stimuli in calf muscle tissue within and between patients with PAD, age‐matched controls, and young healthy individuals.

2. Material and Methods

2.1. Study Participants

This prospective study was based on previously acquired data that was re‐analyzed with a new study aim. The study consisted of three groups of subjects, where patients with PAD and age‐matched controls had been previously included in studies by Törngren et al. [5] and Nordanstig et al. [13], and young healthy individuals were included in a study by Arvidsson et al. [14]. Inclusion criteria for the young control group included age < 40 years and systolic blood pressure < 140 mmHg. For the patient group, inclusion criteria were a confirmed diagnosis of peripheral arterial occlusion disease (PAOD) in Rutherford stage 1–3 with symptom duration > 6 months. For the age‐matched control group, inclusion required normal ABI and an absence of intermittent claudication symptoms. In total, 6 subjects were excluded during the quality control, 3 due to corrupted raw data (2 PAD patients, 1 age‐matched control), and 3 subjects due to insufficient cuffing (2 PAD patients, 1 age‐matched control). While excluded from semi‐quantitative analyses, the appearance of the curves in the subjects with insufficient cuffing were documented for future reference. The clinical and demographic characteristics of the final study group have been summarized in Table 1. It included 17 PAD patients, 8 age‐matched controls, and 14 young controls. Two additional healthy subjects (age: 52, sex: female and age: 34, sex: male) were scanned with altered cuffing schemes to provide supporting results. All subjects gave written informed consent and all procedures performed during the acquisition of these data have been approved by the regional research ethics committee in Gothenburg, Sweden (Dnr 1157–17).

TABLE 1.

Clinical and demographic characteristics for groups PAD patients (n = 17), age‐matched controls (n = 8) and young healthy subjects (n = 14).

Clinical/demographic characteristic	PAD patients (n = 17)	Age matched controls (n = 8)	Young healthy subjects (n = 14)
Age (years), mean ± SD	73 ± 4	68 ± 6	29 ± 4
Male (sex), n (%)	6 (35)	1 (10)	6 (43)
ABI, mean ± SD	0.61 ± 0.12	1.15 ± 0.09	N/A
Hypertension, n (%)	14 (82)	2 (25)	N/A
Diabetes, n (%)	8 (47)	0 (0)	N/A
Smoking per day HSI, mean ± SD	2 ± 0.7	1 ± 1	—
Pain‐free walking distance (m), median (IQR)	102 (30)	N/A	N/A

Abbreviation: HSI, heaviness of smoking index.

2.2. Study Design and Cuff Compression Cycle

Study participants were subject to an 11‐min cuff compression cycle during MRI. The cycle consisted of three phases: a one‐minute baseline followed by a five‐minute arterial occlusion, and finally a five‐minute phase of reactive hyperemia upon deflation (as illustrated in Figure 1). The compression was applied to the upper thigh (as shown in Figure 1), using an automatic tourniquet system (ATS 750, Zimmer, USA) and an MRI‐compatible wide leg tourniquet cuff (ERKA, Berlin, Germany). Inflation and deflation were triggered manually from the MRI control room with applied pressure of 50 mmHg above systolic arm pressure (right arm).

FIGURE 1.

(A) Positioning of leg, pressure cuff and flex coil (top). Cuff compression cycle during dynamic T2* mapping (below), showing the duration of each measurement phase and the time points for cuff inflation and deflation. (B) The first of 207 sets of multi‐echo gradient‐echo acquisitions, where the 11 equidistant echo times range from 2 to 40 ms. (C) $T_2^{*}-\mathrm{map}$ generated from the first set of echoes. (D) Regions of interest covering the soleus (blue), and the lateral (red) and medial (green) gastrocnemius muscles. (E) The descriptive model (gray) fitted to the soleus ROI‐median T2*‐time curve (blue) and the derived perfusion related metrics: ${{\mathrm{T}}_2^{*}}_{\mathit{\text{init}}}$ initial ${\mathrm{T}}_2^{*}$; ${\text{IS}}_{\text{down}}$ initial downslope; posDevLength positive deviation length; posDevMagn positive deviation magnitude; $∆$ UST undershoot transient magnitude; rel. ${T_2^{*}}_{\mathit{max}}$ relative maximum $T_2^{*}$; $∆{T_2^{*}}_{\mathit{max}}$ baseline subtracted maximum $T_2^{*}$; ${\mathrm{HS}}_{\mathrm{up}}$ hyperemic up‐slope; TTHP time to half peak; TTP time to peak; TTHR time to half recovery. Levels and slopes illustrated with arrows, time and level intervals illustrated with double‐headed arrows.

Bedside ABI [12] was measured in all PAD patients and age‐matched controls. For the young control group, MRI data were acquired at an earlier stage before the study protocol was finalized and ABI measurements were not available.

2.3. BOLD MRI Acquisition

The BOLD MRI was performed on a clinical 3 T scanner (Magnetom Skyra, Siemens Healthineers, Erlangen, Germany) using a large 4‐channel receiver flex coil. Subjects were examined lying down in a supine, feet‐first position with the flex coil surrounding the examined calf and positioned near isocenter. $T_2^{*}$‐weighted images were acquired repeatedly in the transverse plane using a multi‐echo gradient‐echo sequence with echo times: 2.0, 5.8, 9.6, 13.4, 17.2, 21.0, 24.8, 28.6, 32.4, 36.2, and 40.0 ms, repetition time (TR): 44 ms, flip angle: 9 degrees, slice thickness: 10 mm, field of view: 160 $\times$ 160 $\mathrm{m}{\mathrm{m}}^2$, acquisition matrix: 128 $\times$ 119, reconstructed image matrix: 128 $\times 128$, parallel imaging acceleration factor: 2, no fat suppression and receiver bandwidth per pixel: 490 Hz. The k‐space trajectory acquired the complete set of echoes within one TR for each k‐space line. Bipolar read‐out gradients were used, so that consecutive k‐space rows were traversed in opposite directions. The sequence was repeated with a temporal resolution of 3.2 s, in total 207 times over the 11‐min cuffing cycle. Cuff inflation and deflation was triggered at the 19th and 114th repetition, respectively.

2.4. Supplemental Measurements

Two supplemental imaging sessions were performed. The first was performed on a healthy volunteer (52‐year‐old female) to investigate the dependence of baseline ${\mathrm{T}}_2^{*}$ levels on breathing type (shallow, deep or normal tidal) using the same BOLD MRI protocol but without cuff application. The second was performed on another healthy volunteer (34‐year‐old male) and involved five repeated cuffing cycles to assess the sensitivity of the automatic tourniquet system to reduced cuff pressure and quadriceps muscle contractions. In these measurements, presented in Supporting Information, the cuff pressure was set to the systolic arm pressure, lower than the pressure used in the patient cohort, to increase the likelihood of cuff leakage.

2.5. $T_2^{*}$ ‐Time Curve Extraction

Regions of interest (ROI) were manually delineated in the soleus, and in the medial and lateral gastrocnemius muscles (as shown in Figure 1D) by a senior medical imaging specialist (JA) with more than 10 years of experience. These were outlined on the first echo image of the first scan repetition using ITK‐SNAP [20]. Based on the 11 echoes in each scan repetition (data shown in Figure 1b), ${\mathrm{T}}_2^{*}$ parameter maps were estimated (Figure 1C) using least squares fit to a negative exponential signal model. To account for spatiotemporal subject movement, the first acquired echo image of each scan repetition was used in a non‐rigid registration scheme using the software Elastix [21], spatially transforming the delineated ROIs (Figure 1D) to each ${\mathrm{T}}_2^{*}$ parameter map. ${\mathrm{T}}_2^{*}$‐time curves were extracted by use of the ROI‐wise median ${\mathrm{T}}_2^{*}$ level from each of the acquired ${\mathrm{T}}_2^{*}$‐maps (Figure 1E).

2.6. Visual Data Survey

Each extracted ROI‐wise ${\mathrm{T}}_2^{*}$‐time curve from soleus and medial gastrocnemius muscles were studied by JA in a blinded data survey. Curves from gastrocnemius lateral were disregarded from analyses based on the high noise level in these curves, which was attributed the substantially smaller size of these compared to soleus and gastrocnemius lateral. Typical healthy BOLD curves include a monotonic decrease in ${\mathrm{T}}_2^{*}$ during the ischemic phase as the cuff is inflated and a fast and high increase in ${\mathrm{T}}_2^{*}$ and subsequent slow return to baseline during the hyperemic response following the cuff‐release [12]. The occurrences of traits that deviated from such an archetype curve were described and the frequency of occurrence noted for each subject group.

2.7. Semi‐Quantitative ${\mathrm{T}}_2^{*}$‐Time Curve Analysis

This study used a semi‐quantitative analysis tool, previously described by Arvidsson et al. [14], to achieve a parameterized signal representation and automatically derive descriptive metrics from ${\mathrm{T}}_2^{*}$‐time curves, as shown in Figure 1E. These include the hyperemic up‐slope (${\mathrm{HS}}_{\mathrm{up}}$), the time to half‐peak (TTHP), the time to peak (TTP) and time to half recovery (TTHR). Both the baseline relative hyperemic peak value (HPV) and the baseline subtracted hyperemic maximum (${{∆T}_2^{*}}_{\mathit{max}}$) were also included for comparison.

After inspection of the individual ${\mathrm{T}}_2^{*}$‐curves, further developments of the semi‐quantitative analysis tool were made to quantify novel curve characteristics. Accordingly, the depth of the undershoot transient (UST) that occurs upon cuff release was determined in ms ($∆$ UST). In addition, deviations from a monotonic decrease in ${\mathrm{T}}_2^{*}$ during cuffing were quantified by the metrics positive deviation length (posDevLength) and positive deviation $\mathit{\text{magnitude}}$ (posDevMagn). These metrics were designed to quantify the total duration and the area under the curve of deviations from a monotonic decrease of the T2* time curve over the time‐period between cuff inflation and cuff deflation. Detailed descriptions of these parameters are provided in Appendix A.

2.8. Statistical Analysis

Non‐parametric 2‐tailed Wilcoxon rank‐sum tests were used to test differences in conventional and novel BOLD metrics between the patient group and the age matched controls. Correlation analyses between BOLD metrics and ABI were performed with datapoints of patients and age‐matched controls pooled using the Spearman's correlation ($\rho$). A p value < 0.05 was considered statistically significant. Statistical analyses were performed in Python version 3.10.14 using SciPy version 1.10.0.

3. Results

3.1. Overview of the ${\mathrm{T}}_2^{*}$‐Time Curves

The initial ${\mathrm{T}}_2^{*}$ level showed high levels of variability between individuals, especially for the PAD patient group and the age‐matched controls, whereas the healthy young individuals showed a lower variability concentrated at the higher end of the parameter range (top row of Figure 2). When normalized to baseline ${\mathrm{T}}_2^{*}$, the visualization of baseline‐relative changes was improved for both intersubject and groupwise comparisons (bottom row of Figure 2). The speed of the hyperemic response, as seen in the initial part of the hyperemic phase showed a distinct difference between the patient group and healthy subject cohorts. Less pronounced differences during the cuff‐induced decline in ${\mathrm{T}}_2^{*}$ (the ischemic phase) were also observed.

FIGURE 2.

Data overview based on ROI‐wise time curves from the soleus muscle in PAD patients (red, n = 17), age‐matched controls (green, n = 8) and young controls (blue, n = 14). Subplots show (A) T2*‐time curves from all subject groups, (B) violin plot covering the initial ${\mathrm{T}}_2^{*}$ values for the three groups, (C) initial ${\mathrm{T}}_2^{*}$ subtracted T2*‐time curves and (D) groupwise median (lines) and IQR (shaded) T2*‐time curves.

The ${\mathrm{T}}_2^{*}$‐time curve in Figure 3A shows curve traits typical for young healthy controls. The response to cuffing included a monotonic decrease in ${\mathrm{T}}_2^{*}$ following the cuff inflation which leveled out after some duration. Even though the ${\mathrm{T}}_2^{*}$ level was not precisely constant during the latter part of the ischemic phase, this part of the curve is hereafter termed the ischemic plateau. Upon cuff‐deflation, the hyperemic response began with a short undershoot transient, followed by a rapid increase (the hyperemic upslope) toward the hyperemic peak. Finally, the curve decreased slowly back to the baseline ${\mathrm{T}}_2^{*}$‐level.

FIGURE 3.

Summary of the data survey. BOLD time curve characteristics from a young healthy subject (A) and patients with peripheral artery disease (B–G). In the first and second columns, curve traits are annotated, while the subsequent columns detail the number and percentage of occurrences in each subject group and muscle.

3.2. New ${\mathrm{T}}_2^{*}$‐Time Curve Traits

The manual data survey is summarized in Figure 3, where curve characteristics found in soleus and gastrocnemius medial muscles (blue and green ROIs in Figure 1) have been illustrated alongside their occurrences in each group. Figure 3, row A, illustrates a representative curve of a young healthy subject including a stable ischemic plateau during cuffing and a large hyperemic overshoot following the cuff release. In row B, observations include unexpected increases in ${\mathrm{T}}_2^{*}$ during cuffing (B1), a trait found in all groups, but more often in patients than controls. A trait that was also seen across groups was a return to baseline level after cuff release that was lower than the initial baseline (B2). The results of supplemental measurements performed to explore the possibility that muscle contractions and different breathing modes could be the driver of such deviations are presented in Supporting Information. Different modes of breathing did slightly affect the ${\mathrm{T}}_2^{*}$ level (Figure S1), however, at a magnitude below the deviations found in the ischemic phase of patient curves. Deviations from a monotonic decrease were quantified by the newly defined metrics posDevLength and posDevMagn (described in detail in Appendix A). Figure 3, row C shows how slow hyperemic responses to cuff release only manifests in the patient group, as can also be seen in Figure 2. Figure 3, row D shows that an absent ischemic plateau and absent hyperemic responses may occur in all groups. These traits could be reproduced in the supplemental measurements by lowering the cuffing pressure to the brachial pressure to increase the risk of cuff leakage as seen in Figure S2. Figure 3, row E illustrates a slight positive transient observed in several measurements during cuff inflation and deflation. However, magnitudes were close to the noise level, limiting the possibility of automated quantification. Figure 3, row F shows that negative curve transients following cuff release could be observed in most young healthy subjects for soleus (n = 12, 86%), with fewer occurrences (n = 7, 78% and n = 8, 42%) for age‐matched healthy controls and PAD patients (Figure 3, F2). Similar results were found also for the gastrocnemius medial ROI. The undershooting cuff‐release transient was assessed by the new metric, $∆\mathit{UST}$. Figure 3, row G, shows an example of the interplay between the dynamics of the soleus and gastrocnemius medial muscles in a PAD patient. The curves (Figure 3G) display an inverse dependency starting from the cuff inflation, throughout the cuff deflation to the hyperemic upslope.

In general, PAD patients produced more complex dynamic ${\mathrm{T}}_2^{*}$ behavior than both the young healthy subjects and the age‐matched control group. As the metric‐based data analysis requires the fitting of a flexible parameterized function, the deviations from the predefined shape disqualified increasing numbers of curves from the gastrocnemius muscle from such analysis (second last row of Figure 3). For the soleus ROI such an analysis was possible for all but three curves that were identified as having non‐successful cuffing (two patients, one age‐matched healthy subject), but only after outlier data points in the ischemic phase of the curve were disregarded for the curve fitting procedure (final row Figure 3). Thus, only BOLD curve metrics from the soleus ROI were derived and compared in further groupwise analyses.

3.3. Groupwise Analysis

The hyperemic upslope was the BOLD curve phase that most clearly differentiated PAD patients from age‐matched controls (Figure 2D and Table 2). Upslope‐related metrics ${\mathrm{HS}}_{\mathrm{up}}$, TTP and TTHP, all produced (top row of Figure 4) significant groupwise differences. These parameters also yielded strong correlations with ABI, with $\left|\rho \right|\ge 0.8$ as shown in Table 2. The speed of the return to baseline, described by TTHR, also differed significantly between patients and age‐matched controls but with larger within group variabilities. The undershooting cuff‐release transient, described by $∆$ UST, showed smaller transients in the patient group, but the overlap with asymptomatic groups was large and the difference was non‐significant (p‐value = 0.15). Significant differences between patients and age‐matched controls were found for metrics describing the hyperemic peak, HPV and $∆{T_2^{*}}_{\mathit{max}}$, but with larger variability and overlaps between groups when compared to metrics describing the hyperemic upslope. The two groups with healthy subjects were checked for groupwise significant differences (Table S2), and only the baseline level, ${{\mathrm{T}}_2^{*}}_{\mathit{\text{init}}}$ (Figure 2B) was significantly different.

TABLE 2.

Mean, standard deviation, groupwise differences, and correlation with ankle‐brachial index (ABI) for each BOLD metric and for groups PAD patients (PAD; n = 17), age‐matched controls (AMC; n = 8), and young healthy subjects (YHS; n = 14).

Metric	PAD patients (n = 17), $\overline{x}$ (sd)	Age matched controls (n = 8), $\overline{x}$ (sd)	Young healthy subjects (n = 14), $\overline{x}$ (sd)	Groupwise mean difference, $∆\overline{x}$ PAD vs. AMC	Groupwise mean difference, $∆\overline{x}$ AMC vs. YHS	ABI‐correlation, $\rho$, pooled data PAD and AMC
ABI (a.u.)	0.61 (0.12)	1.15 (0.08)	—	−0.55*	—	—
${{\mathrm{T}}_2^{*}}_{\mathit{\text{init}}}$ (ms)	22.7 (1.93)	22.1 (3.61)	25.7 (1.31)	0.64	3.66*	−0.1
${\text{IS}}_{\text{down}}$ ${10}^2$ (ms/s)	16.4 (9.92)	4.45 (3.52)	10.3 (20.7)	12.0	5.82	−0.0
posDevLength (s)	62.6 (53.1)	19.2 (31.0)	39.4 (67.4)	43.4	20.2	−0.3
posDevMagn $·{10}^{-3}$ ($s^2$)	13.4 (12.1)	2.83 (4.88)	4.67 (8.29)	10.6	1.84	−0.4
MIV (a.u.)	−0.07 (0.04)	−0.07 (0.04)	−0.07 (0.03)	−0.00	−0.01	0.1
$∆$UST (ms)	0.46 (0.37)	0.94 (0.75)	0.92 (0.56)	−0.48	−0.02	0.2
HPV (a.u.)	1.34 (0.57)	2.28 (0.87)	2.30 (1.46)	−0.94*	0.02	0.3
$∆{T_2^{*}}_{\max }\left(\mathit{ms}\right)$	1.06 (0.03)	1.11 (0.05)	1.09 (0.06)	−0.05*	−0.02	0.3
${\mathrm{HS}}_{\mathrm{up}}$ ${10}^2$ (ms/s)	14.5 (8.80)	77.9 (25.0)	85.1 (35.1)	−63.5*	0.07	0.8
TTHP (s)	45.3 (19.7)	14.4 (2.04)	13.6 (1.41)	30.9*	−0.80	−0.8
TTP (s)	94.9 (30.0)	32.1 (5.72)	29.6 (4.71)	62.8*	−2.48	−0.8
TTHR (s)	59.9 (55.5)	139.3 (53.6)	97.8 (49.3)	−79.4*	−41.5	0.6

Note: Significant differences are marked using *.PAD peripheral artery disease; AMC age matched controls; YHS young healthy subjects; ABI ankle brachial index; ${{\mathrm{T}}_2^{*}}_{\mathit{\text{init}}}$ initial ${\mathrm{T}}_2^{*}$; ${\text{IS}}_{\text{down}}$ initial downslope; posDevLength positive deviation length; posDevMagn positive deviation magnitude; MIV minimal ischemic value; $∆$UST undershoot transient magnitude; HPV hyperemic peak value; $∆{T_2^{*}}_{\mathit{max}}$ baseline subtracted maximum $T_2^{*}$; ${\mathrm{HS}}_{\mathrm{up}}$ hyperemic up‐slope; TTHP time to half peak; TTP time to peak; TTHR time to half recovery.

FIGURE 4.

Ankle brachial index (ABI) and perfusion‐related BOLD metrics obtained via semi‐quantitative curve analyses applied to the soleus muscle in PAD patients (n = 17), age‐matched controls (n = 8) and young controls (n = 14). Metric‐wise display of individual estimates (left), and 2‐sided violin plots (right), grouped such that the estimated histogram profiles of patients are placed to the left and the two control groups are placed to the right in each violin. BOLD perfusion related metrics include ${{\mathrm{T}}_2^{*}}_{\mathit{\text{init}}}$ initial ${\mathrm{T}}_2^{*}$; ${\text{IS}}_{\text{down}}$ initial downslope; posDevLength positive deviation length; $\text{posDevMagn}$ positive deviation magnitude; MIV minimal ischemic value; $∆$ UST undershoot transient magnitude, HPV hyperemic peak value; $∆{T_2^{*}}_{\mathit{max}}$ baseline subtracted maximum $T_2^{*}$; ${\mathrm{HS}}_{\mathrm{up}}$ hyperemic up‐slope, TTHP time to half peak; TTP time to peak; TTHR time to half recovery.

The initial downslope following cuff‐inflation, described by ${\text{IS}}_{\text{down}}$, was not significantly different between patients and age‐matched controls (p = 0.012), as also seen in the median time curves of Figure 2D. The minimal ischemic value (MIV) showed large overlaps between groups (p = 1). Finally, metrics quantifying non‐monotonic decreases during cuffing were not significantly different between groups, with p values = 0.071 and 0.062 for posDevLength and posDevMagn, respectively. Controls displayed low values in most cases and the variability of patients was substantially larger as seen in Figure 4 and in Table 2. Outlier points of high magnitude can be observed for posDevLength, posDevMagn and $\mathrm{I}{\mathrm{S}}_{\text{down}}$ in the young healthy subject group.

4. Discussion

This exploratory study, using a visual survey of $T_2^{*}$‐time curves as well as derived metrics, showed that the variability of curve shapes was larger for gastrocnemius compared to the soleus and larger for the patient group compared to the two healthy subject groups. Groupwise analyses showed that the most discriminant curve characteristics between PAD patients and healthy controls were found in the upslope of the hyperemic response. The hyperemic upslope metrics TTP, TTHP and ${\mathrm{HS}}_{\mathrm{up}}$ were shown to separate PAD patients from age‐matched controls, in accordance with the standard bedside measure, ABI.

The current technique relies heavily on the sensitivity of the MR signal to local concentrations of deoxyhemoglobin within a tissue of interest. Simplified, these concentrations can be influenced by hematocrit and the venous blood volume. In general, hematocrit is viewed as a baseline property, while factors that cause changes to the venous blood volume include oxygen metabolism and blood flow, which in the current experimental design are very much affected by the applied cuffing scheme.

The fact that the two groups are successfully separated by ABI, indicates that macro‐vascular impairment is present in the patient group. Although the hyperemic BOLD response is likely influenced by multiple factors, the initial rise in T2* is primarily driven by the rapid washout of deoxygenated venous blood combined with the inflow of oxygenated arterial blood. These two processes are coupled and determine the shape properties of the early hyperemic response (HS_up, TTP, TTHP). In our data, the upslope metrics consistently outperformed the peak‐based metrics in distinguishing patients from controls. A plausible interpretation is that the rate of T₂*‐increase during the response is associated with the macrovascular blood flow, which is characteristically reduced in PAD. A hypothetical maximum hyperemic peak would occur when the volumes of deoxygenated and oxygenated blood in the tissue simultaneously are at their minimum and maximum, respectively. However, in vivo this rarely occurs. Thus, metrics reflecting the dynamics of the early hyperemic response provide more direct readout of macrovascular limitation in PAD, whereas peak‐based metrics such as the HPV and $∆{T_2^{*}}_{\mathit{max}}$ are more confounded by volume and timing effects, explaining their weaker discriminatory performance.

For the patients, TTP was more than twice as long as for the age‐matched controls (on average ~60 s longer), highlighting the severity in vascular symptoms associated with the PAD disease. While all three upslope parameters may be relevant to consider for future studies, TTP is based on a longer time interval and may thus be suitable for when low temporal resolution imaging is used.

In this study, several new curve traits were also identified and successfully quantified, several showing increased variability in patients when compared to age‐matched controls. This increase in variability is relevant to study further in regards to differentiating phenotypes of PAD. Furthermore, the reproducibility of this imaging technique is relevant to study for patients, especially regarding the novel metrics presented herein.

4.1. $T_2^{*}$ Dependencies Between Calf Muscles

While no blood flow is expected to support the calf muscles during cuffing, several patients showed signs of ${\mathrm{T}}_2^{*}$ dependencies between the soleus and gastrocnemius medial muscles during this phase. This suggests a potential communication between muscle blood pools. Venous drainage of the soleal muscle is primarily provided by the soleal veins but is also supported by the shared sural veins [20]. While the literature appears contradictory regarding the presence of valves in very small veins [21], microvascular valves in human leg muscles have been described as long ago as 1950 [22]. If valve dysfunction is present in intermuscular veins, pressure equalization during cuffing could potentially drive an exchange between venous blood pools. Paramagnetic venous blood traversing from one muscle compartment to another, would manifest as a ${\mathrm{T}}_2^{*}$ decrease in the receiving ROI and an increase in the delivering ROI as observed in the data survey. If these observations and their interpretation prove valid, they may indicate a new means to detect venous valve leakage in peripheral muscle with BOLD MRI.

4.2. Non‐Monotonic Decrease in $T_2^{*}$ During Cuffing

Both in patients with PAD and age‐matched controls a non‐monotonic decrease in $T_2^{*}$ was observed during the ischemic phase of the cuffing procedure. This curve trait was more frequent in patients than in controls. The new metrics posDevLength and posDevMagn did not show any significant groupwise differences. However, the large variabilities found in patients, may reflect physiological differences, potentially useful for separating disease phenotypes. Two types of non‐monotonic decreases were identified: [1] shorter burst‐like increases in ${\mathrm{T}}_2^{*}$, and [2] steady monotonous increases during the latter part of the ischemic phase. If two separate causes drive the increases in ${\mathrm{T}}_2^{*}$ during cuffing, this may explain the large variability in these metrics. Neither a burst‐like increase in ${\mathrm{T}}_2^{*}$ nor a slow increase could be reproduced by contractions in the quadriceps, as shown in supplemental measurements. The automatic tourniquet system effectively regained pressure, yielding only slight changes in ${\mathrm{T}}_2^{*}$, substantially smaller than observed burst‐like ${\mathrm{T}}_2^{*}$‐increases. Slower ${\mathrm{T}}_2^{*}$‐increases were not found to be caused by variations in breathing patterns, as tested in the first supplemental measurement. Both types of ${\mathrm{T}}_2^{*}$ increases during cuffing may however be explained by the previously discussed curve trait since drainage of paramagnetic venous blood from the ROI, would likely manifest as an increase in ${\mathrm{T}}_2^{*}$. Several outlier data points were observed for these metrics, as they are sensitive to small but persisting non‐monotonic declines over the entire course of the ischemic cuffing phase, which can dominate the resulting metric value.

4.3. Absent Ischemic Plateau and Hyperemic Response

Several measured curves (n = 3) were missing both an ischemic plateau and strong hyperemic response, suggesting that arterial occlusion and the accompanying tissue oxygen deficit was not achieved. Additional measurements using a lower cuffing pressure (set at systolic arm pressure) were able to recreate such curves. The additional measurement was repeated with the original (higher) cuff pressure, which reproduced the typical ${\mathrm{T}}_2^{*}$‐time curve with ischemic plateau and hyperemic overshoot. We interpret this as a strong indication that curves without ischemic plateau and hyperemic responses are likely caused by a non‐successful cuffing. A possible explanation is that the selected tourniquet‐pressure, 50 mmHg above brachial arm pressure as suggested by Jacobi et al. [9], may be insufficient to account for variability across patient cohorts. For example, in surgical interventions, the minimum effective midthigh tourniquet pressure has been found to be 90–100 mmHg above brachial arm pressure [23]. With these observations made, we further emphasize the importance of a correct interpretation of observed curve traits in future acquisitions and analyses with the technique.

4.4. The Undershooting Transient Upon Cuff Release

The negative transient in the ${\mathrm{T}}_2^{*}$‐time curve, observed immediately after cuff deflation, was quantified by $∆\mathit{UST}$. $∆\mathit{UST}$ effectively captured this trait, aligning well with its occurrences in the manual data survey. Notably, the transient showed considerable variation and overlap across groups. For soleus, over half of the PAD patients showed no negative transient, compared to < 15% and < 25% in age‐matched and young controls, respectively. While $∆\mathit{UST}$ was not found to be significantly different between groups, such within‐group variations could be interesting to understand better from a physiological perspective, and could help in exploring potential markers for disease phenotyping. Several possible causes of cuff‐induced transients can be considered, including mechanical influences from cuff‐pressure, and resulting dynamical volume‐changes in the arterial and venous compartments, like the Balloon model interpretation of transients in functional MRI (fMRI) [24]. The resemblance to the post‐stimulus undershoot in brain fMRI is noteworthy, as it may reflect transient changes in oxygen metabolism that, upon cuff release, increase deoxyhemoglobin on a faster scale than the exchange of arterial and venous blood [25]. The greater prevalence of this dip in healthy individuals compared to PAD patients may reflect the influence of improved vascular compliance. In a study on cuff‐induced ischemia and reactive hyperemia with a technique called MR‐oximetry, the response to cuffing in the femoral vein and artery was studied [26]. In veins, the technique produced blood oxygenation curves with similarities to those presented in the current study. Specifically, an undershooting transient was identified following the cuff release. This so‐called washout was interpreted as the deoxygenated blood traversing the imaged slice after flow restoration. This interpretation appears plausible also for the undershooting transient upon cuff‐release seen in muscle tissue in this study.

4.5. Future Aspects

Future studies should validate the findings of this exploratory study using comparative statistical analyses and larger datasets. Many gastrocnemius ROI curves in the patient group were unsuitable for metric‐based analysis due to high noise levels, but some gastrocnemius curves also failed to meet basic expectations such as T₂*‐decline during cuffing, and hyperemic peak in T₂* following cuff release. Observed variability in initial T₂* values as well as the atypical curves in gastrocnemius (near subcutaneous fat and the calf's outer boundary) suggest that magnetic field homogeneity and subject positioning are relevant factors to study further. Magnetic field inhomogeneity can vary as the leg slightly moves during cuff inflation. Notably, in gastrocnemius, abrupt positive or negative changes in T₂* were observed during cuffing, possibly due to subtle leg movements and shape changes affecting the magnetic field. The current shimming appears insufficient to fully compensate for these inhomogeneities, why future efforts toward more robust MR image acquisition strategies may be required.

The risk of cuff‐leakage and variabilities related to tourniquet size and thigh girth mismatches could potentially be mitigated by using more recent ATS equipment that automatically adjusts the tourniquet pressure to achieve occlusion at the lowest possible cuffing pressure. Findings that suggest intermuscular venous valve leakages need further validation. While inter‐compartmental fluid exchange in MRI phantoms is challenging, in a controlled setting, the exchange of fluids with different transverse relaxation properties between two compartments during the BOLD MRI scan could serve to reproduce the observed curve dynamics [27].

4.6. Limitations

This study used previously acquired data from a single center using one scanner, thus requiring further validation. The limited inclusion of age‐matched controls and the absence of ABI measurements in young healthy individuals limited comparisons across groups. Although the ABI served as a reference for BOLD metrics, its primary sensitivity to blood flow in large conduit arteries limits the metrics' importance in assessing any microvascular effects that may be captured using BOLD MRI. This study did not include angiographical imaging and is, as such, limited in its interpretation of how specific vascular territories may relate to BOLD curves and derived metrics.

5. Conclusion

This study demonstrated that the upslope of the hyperemic response was the primary curve phase for differentiating PAD patients from age‐matched controls. Newly designed metrics were able to capture observed curve traits, several producing elevated variabilities in PAD patients. Possible root causes of curve traits in patients have been proposed and it has also been shown that an insufficient cuffing‐pressure produces BOLD curves with absent ischemic plateau and weak hyperemic responses, traits whose recognition is important to prevent physiological misinterpretations of ${\mathrm{T}}_2^{*}$‐time curves and derived metrics.

Author Contributions

J.A. and K.L. conceptualized and designed the study. J.A., K.L., S.E., and J.N. acquired the data. J.A., K.L., S.E., J.N., and O.J. analyzed and interpreted the data. J.A. drafted the manuscript. K.L., O.J., S.E., J.N., and E.J. critically reviewed the manuscript.

Funding

The study was financed by Swedish governmental funding of clinical research (ALF). The study was also financed by Sahlgrenska University Hospital funding of research.

Supporting information

Data S1: jmri70204‐sup‐0001‐Supinfo.docx.

Appendix A. Quantification of Non‐Monotonic Decline in T₂ ^* During Ischemia

While the cuff is inflated, consecutive ${\mathrm{T}}_2^{*}$ estimates are expected to decline monotonically to an approximately constant level, the ischemic plateau. Deviations from this expectation, non‐monotonic decreases as illustrated in Figure A1, have been quantified by the two metrics posDevLength ($\rho$) and posDevMagn $\left(\mathrm{\Omega }\right)$. PosDevLength sums the duration of non‐monotonic decrease, while posDevMagn integrates ${\mathrm{T}}_2^{*}$ over the duration of non‐monotonic decrease.

FIGURE A1.

A sequence of non‐monotonic decrease (yellow) of sample‐length 4 illustrating the principle used for derivings the metrics posDevLength (positive deviation length) and pesDevMagn (positive deviation magnitude).

Sequences of non‐monotonic decrease were identified in an iterative fashion, such that a cursor passed through the entire time interval, skipped timepoints already included in a sequence. The metrics posDevLength ($\rho$) and posDevMagn $\left(\mathrm{\Omega }\right)$ were calculated as:

$$j_0=1$$

$$j_{n+1}=\min i:y_i-y_{j_n}<0\bigwedge i>j_n$$

$$\rho =\mathit{dt}\sum _n{j}_{n+1}-j_n-1:{\mathrm{\Phi }}_n$$

$$\mathrm{\Omega }=\mathit{dt}\sum _n\sum _{k=j_n+1}^{j_{n+1}-1}{y}_k-y_{j_n}:{\mathrm{\Phi }}_n,$$

where $n$ denotes the consecutive indices of detected sequences and $i$ denotes the number of samples needed to progress before the signal level again reaches below the initial cursor level, $j_n$ denotes the algorithm cursor, that is the start index of sequence, $n$ and where $\mathit{dt}$ is the sampling interval. To avoid adding noise samples, the sequence must be longer than one sample and the average deviation from the $\text{cursor}-{\mathrm{T}}_2^{*}$ level for the sequence must be larger than twice the noise level:

$${\mathrm{\Phi }}_n=j_{n+1}-j_n>1\bigwedge \left(\frac{1}{j_{n+1}-j_n-1}\sum _{k=j_n+1}^{j_{n+1}-1}{y}_k-y_{j_n}\right)>2\sigma ,$$

where the noise level, $\sigma$, was estimated by averaging the absolute difference between neighboring measurements of the last 20 samples in the cuffing phase:

$$\sigma =\frac{1}{19}\sum _{l=N-20}^{N-1}\left|y_l-y_{l+1}\right|,$$

where N is the measurement index corresponding to the cuff release time point

Arvidsson J., Eriksson S., Jalnefjord O., Johansson E., Nordanstig J., and Lagerstrand K., “Exploring the Dynamics of Ischemia and Reactive Hyperemia With Skeletal Muscle Blood Oxygen Level Dependent MRI in Patients With Peripheral Artery Disease, Age‐Matched Controls, and Young Healthy Subjects,” Journal of Magnetic Resonance Imaging 63, no. 3 (2026): 813–825, 10.1002/jmri.70204.