1. Introduction

Venous thromboembolism (VTE) is a prevalent clinical problem, impacting approximately 1 in 1000 individuals annually in the United States [1,2]. It occurs when an acute blood clot forms in the deep venous system as a result of disturbed blood flow, a hypercoagulable state and/or endothelial injury, as described by Virchow’s triad [1–3]. Numerous risk factors contribute to VTE, including increased age, recent surgery, extended travel, cancer therapy, and pregnancy [2]. VTEs are broadly categorized into deep venous thrombosis (DVT), where blood clots primarily form in the veins of the lower extremities but potentially also in the upper extremities, mesenteric, or cerebral veins, and pulmonary embolisms (PE), where a thrombus from DVT may dislodge and travel to the pulmonary circulation. PEs carry a significant mortality, as one-quarter of patients may experience sudden death [1–3]. DVT itself is not immediately life-threatening, and initially can remain asymptomatic. Unnoticed, the thrombus evolves and continues to grow until the capacity of the venous system is impacted, causing swelling and pain within the affected region [1–3]. Untreated cases may lead to severe outcomes such as neurovascular compromise, limb loss, acute respiratory symptoms, pulmonary hypertension, cardiovascular collapse, thromboembolism, including PE, and potentially death.

Anticoagulation and thrombolytic therapies are the primary treatments for VTE, but their effectiveness decreases as the thrombus evolves [4,5]. While anticoagulation therapy alone may prevent thrombus propagation, recurrence, or PE, it does not always provide an active and complete thrombus resolution and might require additional therapeutic adjuncts. In cases where anticoagulants are contraindicated, patients may undergo the placement of a filter in the inferior vena cava to trap recurring emboli before reaching the lungs [6]. While most VTE cases resolve with treatment, thrombus remodeling may lead to valvular incompetence and post-thrombotic syndrome (PTS), characterized by chronic pain, heaviness, leg cramping, limb edema, stasis dermatitis, and, in severe cases, venous ulcerations [4,7]. PTS has no effective treatment, and its incidence can be as high as 60% at 2 years when acute VTE is treated with anticoagulation therapy alone [1–3]. The public health impact of VTE is further exacerbated by high recurrence rates. Approximately 30% of VTE patients will experience a recurrent episode later in life [1–3].

The high rates of recurrence and post-event complications have driven the development of more targeted treatments, such as minimally invasive endovascular clot removal, including catheter-directed thrombolysis (CDT) or pharmaco-mechanical catheter-directed thrombolysis (PMCDT). PMCDT is primarily performed in patients with extensive or limb/life-threatening PTS. PMCDT involves a catheter-based approach to release thrombolytics, such as recombinant tissue plasminogen activator (rt-PA), directly at the thrombus location, to enzymatically soften and break down the thrombus followed by mechanical force to remove fragmented clots [8]. While some patients benefit from PMCDT [8–10], larger-scale studies report no overall improvement in quality of life [11]. Establishing personalized treatment criteria is advocated to identify subsets of patients likely to benefit from PMCDT [11]. The dynamic nature of thrombosis, with its heterogeneous composition and structural evolution over time, emphasizes the need to recognize thrombus characteristics for optimal management [12].

A thrombus forms through a coagulation process similar to wound healing, involving platelets and fibrin in the initial cascade and eventually recruiting cells such as erythrocytes and white blood cells to create an acute thrombus [13]. As the thrombus ages, smooth muscle cells and fibroblasts are recruited, forming layers of muscle cells and collagenous stroma. The once erythrocyte-rich clot is eventually replaced by collagen and smooth muscle cells, forming a thick layer of collagenous endothelialized tissue known as a chronic thrombus [13]. Within an affected limb, a given thrombus may exist in either an acute, chronic, or combined acute on chronic state, wherein an acute clot forms on the leading edge of pre-existing chronic thrombosis. Thrombolytic drugs, such as rt-PA, target the fibrin- and red blood cell rich structure of an acute clot, and exhibit varied effectiveness based on clot composition [12]. Therefore, the ability to determine the biological state and composition of a thrombus beforehand would greatly aid in making informed decisions about the use of thrombolytic therapy. This is crucial as the treatment poses an increased risk of bleeding complications for patients. For instance, in the extensive clinical study conducted by Vedantham et al. [11], patients received treatment within 6 days from the onset of symptoms. However, this timeframe might not necessarily align with the biological age or structural composition of the thrombus, given the variable asymptomatic nature of DVT.

Herein, we explore the potential of intravascular optical coherence tomography (OCT) imaging to address the clinical challenge of determining the age and composition of a thrombus. OCT is an optical imaging modality which utilizes the backscattering of low-coherence light to produce high-resolution (∼10 µm) tomographic images of tissue microstructure to a depth of 1-2 mm [14–16]. Widely employed in ophthalmology, OCT technology has expanded its applications to include endoscopy, dermatology, and cardiology [16]. Specifically, intravascular OCT (IV-OCT), facilitated by rotating, side-looking probes, as depicted in Fig. 1, has been integrated into interventional cardiology [17–19]. IV-OCT is routinely used in cardiac catheterization procedures by offering high-resolution cross-sectional images of coronary artery morphology, characterizing plaques and guiding stent placement during angioplasty [17–19]. Its form factor would enable direct integration of IV-OCT with endovascular procedures, making it a promising imaging modality to guide catheter-directed thrombolytic therapies.

 

Fig. 1. Flow chart describing imaging / metric extraction pipeline. Thrombi were generated in the IVC of the rats. At 24 hours after thrombus generation, the cohort of “acute” rats were imaged using the lab-made PS-OCT system. The composition of the thrombus at this point would be largely fibrin and coagulated blood. After 28 days, the “chronic” cohort underwent the same process. After imaging, the rats were sacrificed and the veins were formalin fixed, paraffin-embedded and histologically stained with H&E. These sections were then labelled by a pathologist regarding the relative amounts of “acute” and “chronic” clot features in each slice. The histology sections and PS-OCT images were then registered, and the PS-OCT metrics were compared between distinctly acute and chronic sections.

Download Full Size | PDF

To obtain sufficient image contrast for differentiating acute from chronic clots, we furthermore used polarization sensitive OCT (PS-OCT) [20,21] to measure the birefringence of the increased collagen and muscle cell content of chronic clots. Birefringence is an optical property which manifests when different polarization states experience distinct indices of refraction, due to the structural anisotropy of the underlying material. Both collagen-rich fibrous tissues and the smooth muscle cell layer in the vascular wall exhibit pronounced birefringence. Our team has made significant progress in the development and validation of intravascular polarization-sensitive OCT (IV-PS-OCT) [22–25], making this contrast available in a catheter-based setting. In addition to depth-resolved birefringence, the degree of polarization (DOP) measures how the tissue randomizes the detected polarization states. Multiple scattering of light or propagation through tissues with birefringent structures, which vary on a spatial scale smaller than the OCT resolution, leads to a reduced DOP. Given that collagen and smooth muscle content serves as a primary distinguishing factor between acute and chronic thrombi, these polarization metrics could serve as a potentially valuable and impactful metric for assessing the composition and age of DVT.

Considering the alterations in the structural collagen composition throughout the aging process of a venous thrombus and the proven sensitivity of IV-PS-OCT to these changes, we propose that PS-OCT could emerge as a valuable diagnostic tool when integrated with CDT or PMCDT interventions. The present work represents the first investigation to validate the use of PS-OCT for assessing the thrombus age in-vivo using a rat model of acute and chronic DVT. Using this model, we employed IV-PS-OCT to visualize the birefringence and the degree of polarization (DOP) of both acute and chronic thrombi, developing metrics based on these properties to effectively distinguish between the two thrombus types.

2. Methods

2.1 Catheter-based PS-OCT system

PS-OCT imaging was performed using a custom-built optical frequency domain system as outlined in previous publications [23,25]. Briefly, our system utilizes a laser source that was swept across a 115 nm range centered at 1320 nm scanning at a rate of 54 kHz. Considering an estimated tissue refractive index of n = 1.33, we determined the axial full-width half-maximum of the point spread function to be 9.3 µm. For imaging, we used custom-made 2.7 Fr intravascular catheters, originally designed for use in human coronary arteries [24] and adapted for use in rats by removing the guide-wire provision. The side-facing imaging probe was connected to the imaging console through a rotary joint, which rotated and translated backward within a transparent sheath, creating a helically scanned focus of ∼30 µm in diameter. The catheter pullback speed resulted in a pitch of 20 µm between adjacent cross-sections, with each cross-section containing 2048 one-dimensional depth scans (A-lines). To enable polarization sensitivity, we employed an electro-optic modulator that alternated the polarization state of the light in the sample arm between two orthogonal states on the Poincaré sphere for adjacent A-lines. The information obtained from input state modulation combined with polarized-diverse detection served to extract the polarization properties of the sample.

2.2 PS-OCT signal processing

The acquired raw PS-OCT data was processed using a custom MATLAB-based algorithm detailed in previous works [22,23]. The reconstruction employed spectral binning to estimate the polarization mode dispersion (PMD), or wavelength-dependent retardance present in the catheter-based system [22]. The acquired measurements were then corrected for PMD by compensating with the estimated system effects, before deriving the local, depth-resolved birefringence. Combined, we obtained three distinct and complementary signals: 1) intensity, highlighting scattering tissue structures, shown in Fig. 3(a), 2) birefringence, visualized in Fig. 3(b), and 3) degree of polarization (DOP), shown in Fig. 3(c). The DOP metric used in this work is described with the following expression:

2.3 In-vivo experimental design

All experimental animal procedures were approved by the institutional animal committee IACUC and conform to the Guide for the Care and Use of Laboratory Animals published by the US National Institutes of Health (NIH Publication No. 85-23, revised 1996). To investigate the potential utility of PS-OCT for characterizing thrombus structures, we conducted experiments at two distinct time points in an in-vivo rat model of DVT, aiming to capture both acute and chronic thrombi stages. To induce a thrombus in the inferior vena cava (IVC) (Fig. 1), 7 male Sprague Dawley rats from Charles River Laboratories, Boston, MA, were used. The rats were anesthetized with 2% isoflurane in 100% oxygen, delivered via a face mask. The animals were then placed in a supine position on a warming platform to maintain a core temperature of 37°C, measured with a rectal probe using the ATC1000 system (World Precision Instruments, Sarasota, FL). The abdominal wall was shaved and disinfected using repeated betadine application. A midline laparotomy was performed to access the abdominal cavity, and the IVC was exposed. The IVC was dissected away from the aorta through a combination of sharp and blunt dissections and permanently ligated immediately inferior to the renal veins using a 5-0 silk ligature. To promote venous stasis and subsequent thrombus formation, all side branches of the IVC, from the renal veins to the common iliac veins, including the lumbar and gonadal veins, were ligated. Following the IVC ligation, the peritoneum, muscle layers, and skin were approximated in layers using 5-0 running sutures for the deep layers and subcuticular sutures were used to close the top skin layer. The rats were then randomly divided into two groups: an acute group with clots of acute thrombus composition imaged and subsequently sacrificed at 24 hours (n = 3), and a second group imaged at 28 days after the IVC ligation procedure followed by euthanasia (n = 4) (Fig. 1). To assess PS-OCT in a normal vessel, one rat without the induction of thrombus was also imaged, serving as a control.

To perform IV-PS-OCT imaging of the IVC thrombus, a surgical incision was made to expose the femoral vein in the inguinal region, facilitating the introduction of a catheter. Subsequently, the IVC was exposed to visually track the catheter tip within the ligated vein. The catheter tip was advanced and manipulated proximally until reaching the ligation site in the IVC. Employing the IV-PS-OCT system, the entire IVC was imaged from the ligature to the common iliac veins bifurcation, covering a distance of approximately 3.5 cm and resulting in over 1500 IV-PS-OCT cross-sectional images per animal. Following imaging, the rats were euthanized, and venous tissues were harvested, formalin-fixed, and processed for histological evaluation.

2.4 Histology registration and polarization metric extraction

The IV-PS-OCT polarization images and H&E histology sections underwent manual registration. The rat vein tissues were divided for paraffin embedding into 1-2 cm “blocks” resulting in 3-4 blocks per rat. These blocks were further sectioned into 5 µm thick histological cross-sections, spaced 1 mm apart, generating about 25 sections per animal. While this yielded a reliable intra-block section offset, a variable inter-block offset required adjustment. Anatomic features, like blood vessels, nerves, and lumen size, were used for block registration to IV-PS-OCT data, necessitating at least one clearly registered section per block (see Fig. 2).

A pathologist (L.P.H.), who was blinded to the animal cohorts and the IV-PS-OCT images, classified each histologic cross-section with identifiable presence of thrombus as either “acute” (n = 73) or “chronic” (n = 41). The presence of crystalized blood was also noted, which is a known artifact specific to blood in rats [26]. Histological sections and corresponding matched PS-OCT images without identified clot were excluded from the analysis (n = 45).

Intensity, birefringence, and DOP signals were analyzed by masking the catheter and calculating their cross-section-wide median values across all pixels with a DOP above a threshold of 0.7 and comparing them to their respective histological labels. Notably, perivascular muscle signal (see Fig. 2(c)) was consistently excluded from median computation whenever visible.

 

Fig. 2. Registration process and feature matching. a) A graphic representation of the registration of IV-PS-OCT images and histology. Representative pairs of IV-PS-OCT birefringence cross-sections (1) and histology slices (2) are shown for acute (b) and chronic (c) thrombi. Areas of muscle are shown in green and external blood vessels are highlighted in cyan. Scale bar: 1 mm

Download Full Size | PDF

2.5 Statistical analysis and linear discriminant model

A two sample Kolmogorov-Smirnov (KS) test was used to assess the statistically significant differences between the PS-OCT cross-section metrics. This method was chosen due to its generality and non-parametric nature, as the underlying distribution of these metrics are unknown. If the determined p-value from the KS test was less than the significance level of 0.05, the two distributions were considered statistically significantly different.

A linear discriminant analysis (LDA) model was developed to improve the classification ability of PS-OCT. The inputs for the LDA model were the median values for a) intensity, b) retardance and c) DOP for all pixels where DOP > 0.7 across the matched cross-sectional image. This resulted in one value for each metric (intensity, birefringence and DOP) per cross-section. Cross-sections of normal vein without induced thrombosis were removed from the LDA classifier, due to the lack of differentiation ability between coagulated and normal blood in this specific study. A k-fold validation technique was implemented where each training round began with the input data from one rat in each category (acute and chronic) being reserved for independent, de novo testing after training with the rest of the input data from the remaining 5 rats. The training process and testing process was then repeated for each possible testing pair of one chronic rat and one acute rat (12 rounds in total), and the average taken to produce the final displayed LDA model.

3. Results

3.1 Separation of acute and chronic clot sections

After histological registration, there were a total of 41 matched cross-sections which were labelled as chronic and 73 matched cross-sections which were labelled as acute. There were also 43 cross-sections used from the control rat to represent normal vasculature. Qualitative comparisons of IV-PS-OCT images from acute, chronic, and control rats are shown in Fig. 3. Chronic thrombus segments displayed notably larger amounts of highly birefringent tissue compared to both acute and normal segments (Fig. 3(b).4). When comparing median polarization property values between labeled acute and chronic clots, moderate, yet statistically significant, separation was found with each metric (Fig. 4). The median intensity metric (Fig. 4(a).1) was increased in the chronic thrombus group compared to both normal and acute clots, which is hypothesized to be due to the increased amount of highly scattering fibrous tissue in the vessel wall. The median birefringence was increased in the chronic clots, which is also expected due to the increase in collagenous content within chronic clots compared to acute clots. Finally, the DOP signal (after thresholding) was observed to be greater in acute and normal sections as compared to the chronic thrombus segments.

 

Fig. 3. Representative IV-PS-OCT images of a) acute clot, b) chronic clot and c) normal rat vein. The intensity (1), birefringence (2) and degree of polarization (3) metrics are shown in cross-sections to demonstrate the qualitative difference between clot types. Longitudinal birefringence en-face images in the pullback direction (4) present the variation of the birefringence signal 170 µm outside the sheath. The depth used to generate en-face images is denoted by white dashed circles (a.2, b.2 and c.2). White dashed lines present in the en-face view (a.4, b.4 and c.4) denote the positions of the representative cross-sections (a.1-a.3, b.1-b.3 and c.1-c.3). Horizontal scale bars: 1 mm. Vertical scale bar: 5 mm.

Download Full Size | PDF

 

Fig. 4. Results of the analysis of PS-OCT metrics. a) Boxplots of raw PS-OCT metrics when registered with histology labels. Dots indicate individual median cross-section values for chronic (blue), acute (red) and normal (green) labelled matched histology. Each hue of the dots represent a different individual rat. b) Observed separation plane of LDA model separating acute (red) and chronic (blue) sections. c) ROC curves of all raw metrics + LDA model. d) Sensitivity, specificity, and accuracy of raw metric classifier and LDA model, with cutoff values determined by optimizing overall accuracy. In this table, Biref indicates birefringence, DOP indicates degree of polarization, and LDA indicates the results from using the linear discriminant analysis model.

Download Full Size | PDF

Using the two sample KS test, each metric showed statistically significant differences in all three median metrics between chronic and acute clots. However, as observed in both the receiver operator curve (ROC) plot (Fig. 4(c)) and the summary of classification ability (Fig. 4(d)), the ability to correctly classify clots as acute or chronic was limited using these metrics alone. Sensitivity is defined as the ability of an individual metric to correctly identify chronic clots and is the ratio of total number of correct identifications of chronic clots over the total number of chronic clots present. The specificity of the classifier describes the ability of the metric to classify acute clots. This metric is defined as the total number of correctly identified acute clots over the total number of present acute clots. Lastly, the accuracy of each metric is determined by dividing the total number of correct identifications, combining both chronic and acute correct classifications, over the total number of samples. The ROC curve presents the classification ability of each metric as a tradeoff of false positive rate versus true positive rate. The horizontal axis indicates the false positive rate while the vertical axis demonstrates the true positive rate of the metric. An ideal classifier would have a true positive rate of 1 and a false positive rate of 0 (indicated by the top left corner of the ROC plot). The birefringence metric performed the best out of the three independent metrics, with an overall accuracy of 91.1%, and a sensitivity of 82.9%. The intensity metric was able to classify clots with an overall accuracy of 78.8%, at a significantly lower sensitivity of 63.4%. Lastly, the DOP metric classified clots with an accuracy of 83.2%, however with the poorest sensitivity of all three metrics at 58.6%. In order to improve the classification power of the technique, we combined the three metrics by using linear discriminant analysis (LDA).

3.2 LDA model results

The average accuracy, sensitivity and specificity of the final LDA model are presented in Fig. 4(d). The standard deviation for the sensitivity, specificity and accuracy over all trials used to generate the final LDA model was 2.7%, 5.1% and 3.9%, respectively. Inspecting the cross-sections which performed poorly, they frequently correspond to vessels of chronic thrombus with large lumen diameters, reducing the visualization of the vessel wall. The plane generated by the LDA model in Fig. 4(b) shows the separation between the acute (red) and chronic (blue) PS-OCT metrics in 3D space, where the axes of the 3D space are defined by the intensity, birefringence and DOP metrics. Each labeled cross-section represents a point on the 3D graph, with the acute points shown in red and the chronic clots shown in blue. The semi-transparent black plane is comprised of the 3 linear components of the LDA model. The direction orthogonal to the plane represents the axis achieving best separation, avoiding the overlap present in each individual metric. The normalized vector to the plane was determined to be [-0.73, -0.07, 0.68], where the first dimension indicates the intensity proportion, the second dimension indicates the birefringence proportion, and the third metric indicates the DOP proportion included in the LDA classifier. The ROC curve for this combined metric is shown by the green curve in Fig. 4(c). The overall accuracy for the LDA model was 98.2% with a sensitivity of 97.6% and a specificity of 98.6%.

4. Discussion

DVT remains a substantial public health concern due to its elevated recurrence rates, morbidity, and the associated risks of mortality, particularly from pulmonary embolism. The precise determination of thrombus age remains of paramount importance in the management of DVT patients. However, current clinically utilized imaging tools for aging thrombi are suboptimal, necessitating the development of more advanced imaging technology. A refined approach, capable of accurately determining the age of deep vein thrombi (acute versus chronic) based on confirming structural characteristics with high precision, has the potential to significantly impact decision-making and optimize the treatment of patients with DVT.

Using an in-vivo rat model of DVT, we have successfully demonstrated the capability of IV-PS-OCT in accurately differentiating between acute and chronic (aged) venous thrombus. The results indicated a high level of accuracy (98.2%), sensitivity (97.6%), and specificity (98.6%). This finding holds significant implications, particularly in the context of assessing the potential effectiveness of catheter-directed thrombolytic therapies. The precise differentiation between acute and chronic venous thrombi is crucial due to the well-known refractory nature of chronic, collagenous clots to thrombolytic therapy. Avoiding potentially risky thrombolytic treatments is essential to mitigate the risk of hemorrhagic complications for the patient [11,12].

IV-PS-OCT generates three imaging metrics of complementary contrast: intensity, birefringence, and DOP. For each registered cross-section, the median value for each contrast was calculated. Each metric demonstrated statistically significant differences between acute and chronic but lacked the separative power to classify clots with high accuracy (78.7%, 91.1% and 83.2%, for intensity, birefringence and DOP, respectively). The intensity metric displayed a lower accuracy compared to both the DOP and birefringence metric. This is likely due to scattering being the primary contrast mechanism for the intensity metric. Both the collagenous chronic clots and the red blood cell-rich acute clots display increased scattering, leading to poor differentiation between the two clot types. The DOP metric, on the other hand, displayed a very poor sensitivity for chronic clots (58.6%), yet excellent specificity (97.2%). This is thought to be due to the broad range of DOP values found in chronic clots, reflecting their heterogenous nature, yet a relatively narrow range of DOP values in acute clots. Each metric holds independent information, allowing for the combination of the three metrics in linear discriminant analysis (LDA), which allowed for improved classification ability. Interestingly, when looking at the normal vector describing the LDA plane, a significant amount of the LDA classification ability relies on both the intensity and DOP metric, even when the individual metrics of intensity and DOP perform poorly independently. This again illustrates that the combination of PS-OCT metrics is very powerful in differentiating these clots.

IV-PS-OCT presents a promising complementary addition to CDT and PMCDT for the treatment of DVT. It would readily integrate into the existing catheter-based therapy workflow and offered high accuracy based on this preliminary investigation. This work also highlights the utility of IV-PS-OCT methods compared to conventional IV-OCT for obtaining morphological information in DVT, such as collagen content. The intensity metric is the only available information with conventional IV-OCT. As seen in Fig. 4(d), the intensity metric had a maximum accuracy of 78.7%, with a 63.4% sensitivity for chronic clots, substantially inferior to the classification achieved with the polarization metrics.

One of the limitations of this study is specific to the size of the rat vein and the PS-OCT catheter. Conventional IV-OCT catheters for use in human vessels have a guide-wire provision and are deployed through a guide catheter, which enables flushing with radiopaque contrast agent or saline to displace blood and obtain a clear view on the vessel wall. Rat veins are too small for guide-wire-based deployment, and thus, no flushing was possible. Because coagulated and flowing blood have similar scattering signals, PS-OCT was unable to differentiate between acute clots and normal vessels containing blood in this study due to this animal model specific limitation. Analysis of the dynamic signal properties, which are expected to greatly vary between coagulated and fresh blood, may offer a possible further refinement for improved analysis in this setting. Additionally, adopting PS-OCT into DVT clinical practice would require new catheter development or repurposing of existing cardiac catheters which would require regulatory approval.

A major limitation of this study is the small sample size of 8 rats, which limits the generalizability of these results. As a proof of concept, the primary goal of this work was to determine whether PS-OCT would have any sensitivity for differentiating chronic from acute thrombi, before building on these results in future work. With the promising results derived from this small cohort study, upcoming studies will expand upon these findings by adding larger animal models that better model the human venous system, and include an increased number of animals and timepoints to better determine the sensitivity of PS-OCT to small changes in thrombus structure.

In conclusion, this work demonstrates the first in-vivo study of deep venous thrombosis structural and compositional assessment using catheter-based IV-PS-OCT. Our findings revealed statistically significant distinctions in the polarimetric signals between acute and chronic clots. By integrating these metrics into a linear discriminant model, we achieved a classification accuracy of 98.2% for distinguishing between acute and chronic clots. These outcomes affirm the sensitivity of IV-PS-OCT to detect structural and compositional variations in thrombi, crucial factors influencing therapeutic effectiveness. This sensitivity suggests a potential role in guiding advanced thrombolytic therapies in future clinical applications. Future work will focus on advancing the translation of IV-PS-OCT imaging to larger animal models with larger vessels, and on incorporating additional time points to ascertain the technique's sensitivity to more subtle changes in thrombi structure. These efforts collectively aim to enhance our understanding and application of IV-PS-OCT in the context of venous thrombosis, paving the way for more refined and effective clinical interventions.