1. Introduction

Thrombus formation is a physiological response to prevent excessive blood loss when a blood vessel is damaged. When a vascular injury occurs, platelets become activated and interact with the extracellular matrix underneath the vascular endothelium, causing them to adhere to the injured vessel. The activated platelets form a stable plug held in place with fibrin. Under hemostatic circumstances, thrombi are dissolved during fibrinolysis and cleared by the body [1,2]. However, coagulation disorders, endothelial dysfunction, and hyperinflammation can affect the normal formation or clearing of a thrombus, thus increasing the risk for severe health-threatening diseases such as deep vein thrombosis [3,4], pulmonary embolism [5], stroke [6,7], and heart attack [8,9]. Consequently, understanding, parameterizing, and measuring the formation of thrombi is essential for developing preventive and therapeutic strategies for many cardiovascular and hematologic disorders.

Several techniques focusing on different physical properties of blood have emerged to monitor thrombus formation [10]. These techniques include electrical [11–13] and electrochemical [14] signals, electro-mechanical analysis [15–17], photoacoustics (PA) [18,19] and optical coherence tomography (OCT) [20–22]. Although most methods provide precise assessments, one of their main limitations lies in requiring a static state or non-physiological flow of the blood. Therefore, they lack applicability to detailed studies of thrombus formation, where the shear wave velocity, vessel size, and hematocrit modify the clotting process [20,23]. Due to their non-invasive and imaging capabilities, PA- and OCT-based methods that can analyze flowing blood have become attractive approaches.

In PA, an acoustic transducer detects acoustic waves produced by the tissue after excitation with a pulsed laser. In the case of forming thrombi, the changes in absorption are detectable in the PA signal, revealing its composition and age [18,19]. However, PA application remains restricted due to limited spatial and temporal resolution and high background noise. Moreover, monitoring blood clotting with OCT is growing in attention due to its potential application in vivo and capability to characterize properties of whole blood non-invasively, such as hematocrit [24], blood oxygen saturation [25], blood fibrinogen level [26], and total hemoglobin concentrations [27]. OCT uses backscattered light to create high-resolution images of tissue [28]. In the case of blood, the changes in viscoelastic properties during thrombus formation directly influence the absorption and scattering experienced by light [22,23]. Therefore, monitoring the dynamics of thrombi formation based on the OCT signal over time has been demonstrated.

By evaluating the decay of the OCT signal in clotting blood, Xu et al. have investigated the clotting time under static and dynamic conditions for different blood flow velocities, hematocrits, tube sizes, and fibrinogen levels [20,23,26]. Furthermore, the authors developed an optical coherence elastography (OCE) method to obtain viscoelastic properties of blood by estimating shear wave velocities and mechanical displacements [21]. However, OCE requires specialized equipment to induce mechanical waves with quantifiable movements. More recently, Tang et al. proposed a correlation-based technique to determine the coagulation time, thrombus formation duration, and thrombus strength from the autocorrelation of the OCT signal at different clotting stages [22]. However, these works have characterized blood coagulation by analyzing the behavior of consecutive A-lines of flowing blood through capillary glasses, thus imaging forming thrombi inside the blood flow has not been achieved. The difficulty in imaging and quantifying thrombus formation arises from the challenging differentiation and localization of developing thrombi in free-flowing blood based on the OCT signal. This limitation is magnified due to a lack of adherence of the thrombi to the capillary glass, thus forming thrombi being flushed from the imaging area. We aim to overcome those limitations by combining spatial and temporal analysis of the correlation of the OCT signal while imaging thrombi formation in clotting blood. The blood flows through a vessel-on-chip (VoC) device that provides a human blood vessel model with cellular-level interaction between blood and endothelial cells.

VoC devices are micro-physiological systems designed to mimic human vessels by culturing human vascular cells in a spatially controlled, dynamically perfused microenvironment. [29–36]. VoC models are used for studying vascular inflammation and employ blood perfusion assays to unveil the underlying mechanisms of these processes [37–45], thus thrombus formation has been previously simulated [46–48]. The dynamics of the formation of a thrombus rely on the interplay of many parameters, such as platelet behavior [49,50]. Real-time information about the dynamic formation of a thrombus can facilitate understanding thromboembolic events that occur under different types of flow, thus easing anti-thrombotic drug development [51–53]. Using fluorescence microscopy, VoCs can reveal the aggregation of thrombus components, such as platelets and fibrin, during and after active blood perfusion. However, fluorescence microscopy is a limited read-out, as it relies on selected markers. Furthermore, these techniques only provide insight into two-dimensional end-point read-outs. Therefore, three-dimensional, non-invasive, and real-time monitoring methods such as OCT present an opportunity to understand the dynamics of thrombus formation in these VoC devices [54,55].

In this work, we combine OCT with a microfluidic VoC device to monitor thrombus formation in real time inside flowing blood. The VoC device mimics the interaction between endothelial cells and human whole blood, thus allowing the localization of thrombi. Thrombi formation inside the VoC is imaged with a visible-light OCT (vis-OCT) system during a blood perfusion assay. We develop a method to monitor the progression of the thrombi based on the correlation of the OCT signal. The correlation analysis can follow the development of thrombi throughout the duration of the perfusion assay, and the differences in the correlation indicate the formation and growth of thrombi inside the blood flow. We validate our results with microscopy images of labeled fibrin and platelets after the blood perfusion assay. This OCT thrombus formation technique can be used to observe thrombus formation on other VoC devices, where multidimensional thrombus formation observations can lead to a better understanding of clotting disorders and other vascular diseases.

2. Materials and methods

2.1 Microfluidic vessel-on-chip device

2.1.1 Microfluidic chip fabrication

The microfluidic chip consists of 4 channels which are $1$ cm in length and $300 \times 52$ $\mathrm{\mu}$m in width and height (as shown in Fig. 1(a) and Fig. 1(b)). The chip was fabricated by polydimethylsiloxane (PDMS, Sylgard 184, Mavom, the Netherlands)-based soft lithography using a wafer, which was produced by SU-8-based lithography using designs made in CleWin (WieWin software, the Netherlands) [34,37]. PDMS was added to the wafer at a ratio of $10:1$ (base: crosslinker), left to cure overnight at $65^\circ$C, and removed after. Inlet and outlet holes of $1$ mm were created employing biopsy punchers (RBP-10P, Robbins Instruments, USA). PDMS was spin-coated (SPS Spin150, SPS-international, The Netherlands) on a glass microscope slide (with thickness $\sim 150$ $\mathrm{\mu}$m), and it was left to cure overnight at $65^\circ$C (Fig. 1(a)). The surfaces of the PDMS chip and the microscope slide were activated by exposing them to air plasma ($50$ W) for $40$ s (Cute, Femto Science, South Korea), after which the microfluidic chip was bonded to the microscope slide, as displayed in Fig. 1(a). The activated chip was then used for cell seeding (Fig. 1(b)).

 

Fig. 1. Overview of blood perfusion assay using OCT imaging. (a) The microfluidic chip is made from PDMS and contains 4 straight channels assessable separately. (b) The microfluidic chip consists of a $1$ cm long channel with $300\times 52$ $\mathrm{\mu}$m width and height. These channels are seeded with HUVECs and after 24 hours a monolayer has formed, then blood can be perfused through the channels. (c) Overview of blood perfusion set-up when imaging using OCT. (d) Brightfield image of the quiescent cell monolayer. (e) Fluorescent microscopy image of the monolayer distinguishing nuclei in blue and F-actin in green. (f) After the blood perfusion assay thrombus markers such as fibrin (in purple) and platelets (in green) are imaged using a fluorescent microscope. (g) Schematic illustration of the vis-OCT set-up used to monitor thrombus formation. L: lens, SC: supercontinuum source, F: low-pass filter, NDF: neutral density filter, BS: beam splitter, DC: dispersion compensation element, M: mirror, GS: galvanometer scanner, SL: scanning lens.

Download Full Size | PDF

2.1.2 Cell culturing and seeding

Human umbilical vein endothelial cells (HUVECs LO CC-2519, Westburg, The Netherlands) were cultured in pre-coated collagen culture flasks (#690950, Greiner Bio One2, The Netherlands) at $37^\circ$C and $5$% CO$_2$ using endothelial cell growth medium 2 (EGM-2, #C-22111, PromoCell inc., Germany). Immediately after plasma activation, the chips were coated using a $0.1$ mg/ml collagen solution (Collagen type-I, rat tail #A1048301, Thermofisher, The Netherlands). The coating was incubated at $37^\circ$C and $5$% CO$_2$ for $1$ hour. The coated chips were flushed with EGM-2 and prepared for cell seeding. HUVECs were removed from the cell culture flask by incubation with $3$ ml Trypsin-EDTA (#15400054, Thermofisher, The Netherlands) for $3$ minutes at $37^\circ$C and $5$% CO$_2$. A cell culture medium in a $1:2$ ratio was added to the flask to inactivate trypsin. The cells were centrifuged for $5$ minutes at $390 \times $g and counted using a brightfield cell counter (Lunatrade; automated cell counter, Logos Biosystems, France). The cells were seeded on the top and bottom of the channel at $15\times 10^6$ cells/ml for at least $30$ min for HUVECs to attach (Fig. 1(b)). A total of $150$ $\mathrm{\mu}$l of EGM-2 were added to the microfluidic channels, and the chips were placed on a rocking table (BenchBlottertrade; 2D platform rocker, Benchmark Scientific, USA) inside the incubator for $24$ hours, after which a perfusable monolayer was formed (Fig. 1(b) and Fig. 1(d)). To observe the endothelial monolayer, F-actin and DNA in the cells were stained with Phalloidin-A488 and DAPI (R37110, R37605, Thermofisher, the Netherlands) as depicted in Fig. 1(e).

2.1.3 Blood perfusion assay

The Experimental Centre for Technical Medicine (ECTM, Techmed Centre, University of Twente) provided human whole blood samples. The local medical research ethics committee (METC Twente) approved the blood collection procedure. In agreement with the Declaration of Helsinki, all volunteers gave written informed consent. Whole blood samples were collected in vacuvette tubes containing $3.2$% citrate (#455322, GreinerBio, The Netherlands). The samples were used within 4 hours after the blood draw. The first tube was discarded due to the risk of contamination with tissue thromboplastin during the blood draw [56]. Then, platelets and fibrin were stained with CD41-PE (1% (v/v), #MHCD4104, Thermo Fisher, The Netherlands) and fluorescein-conjugated fibrinogen (Fibrinogen from human plasma, #F13192, Invitrogen, The Netherlands) respectively for 10 minutes. A bent $14$ gauge blunt needle replaced the outlet pipette tip and connected the chip through tubing (Tygon, Fishersci, The Netherlands) to a syringe pump (Harvard PHD 2000, Harvard Apparatus, USA). Immediately before the perfusion assay, blood was recalcified using a recalcification buffer containing HEPES (15630080, Thermofisher, The Netherlands), $63.2$ mM CaCl$_2$ ($1$ M stock, #21115, Merck, Germany), and $31.6$ mM MgCl$_2$ (#J61014.AK, Thermofisher, The Netherlands). During blood perfusion, the pump pulled blood through the channel at $7.8$ $\mathrm{\mu}$l/min for $\sim 15$ minutes, as shown in Fig. 1(c). After that, the channels were flushed with EGM-2 and fixated with $4$% formaldehyde for $15$ minutes. Finally, an EVOS M5000 imaging system (Thermo Fisher, The Netherlands) was used to obtain microscopy validation images at $4\times$ magnification, as shown in Fig. 1(f).

2.2 Optical coherence tomography

2.2.1 OCT system

We use a custom-built OCT system optimized for visible light, described in detail in our previous work [27,57] and schematized in Fig. 1(g). Briefly, light propagates from a supercontinuum broadband source (SC, SuperK EXTREME EXB-6, NKT Photonics, Denmark) through three lenses (L1: LD2746-A, L2: LD2060-A, L3: LB1471-A, Thorlabs, USA) that expand and collimate the beam. A short-pass filter (FESH0700, Thorlabs, USA) filters out all wavelengths above 700 nm before the light arrives at the open-air Michelson interferometer. In the interferometer, a $10:90$ beam splitter (BS028, Thorlabs, USA) reflects $90$% of the incoming light towards the reference arm and transmits $10$% towards the sample arm. In the reference arm, a variable neutral density filter (NDC-50C-2M-A, Thorlabs, USA) controls the intensity of light propagating through a dispersion compensation block composed by a microscope slide equal to the one on which PDMS was molded (Section 2.1.1) and a lens-equivalent dispersion compensation glass (LSM03DC-VIS, Thorlabs, USA). Then, the reference mirror (PF10-03-P01, Thorlabs, USA) reflects the beam to the beam splitter. In the sample arm, a variable neutral density filter (NDC-50C-2M-A, Thorlabs, USA) attenuates the light propagating to a galvanometer scanner (8320K, Cambridge Technology, USA) that steers the beam. After the scanner, a scanning lens (LSM03-VIS, Thorlabs, USA) with a $39$ mm focal length focuses the beam at the channel of the VoC device. The $1/e^{2}$ diameter of the focused beam is $\sim 12$ $\mathrm{\mu}$m. Note that the system images the VoC device from the bottom, where the microscope slide is located (Fig. 1(c)). Backscattered light from the VoC and light reflected at the reference arm combine at the beam splitter and is collected by a single-mode fiber (S405XP, Thorlabs, USA). The fiber guides the light into a custom-built spectrometer (HoloSpec f/1.8i, Kaiser Optical Systems, USA), where a grating disperses it on a line camera (Sprint spL4096-140km, Basler, Germany). The camera has an exposure time of $58$ $\mathrm{\mu}$s at a line rate of $16.7$ kHz for a delivered power at the sample of $\sim 5$ mW. The spectrometer has a spectral resolution of $0.1$ nm within $480-660$ nm. The collected spectrum has a full width at half-maximum of $\sim 75$ nm centered at $555$ nm, leading to a theoretical axial resolution of $\sim 1.8$ $\mathrm{\mu}$m in air.

2.2.2 Imaging protocol

At the beginning of the blood perfusion assay, VoCs were placed in the OCT setup with the bottom of the microscope slide facing towards the system (Fig. 1(c)) and were imaged continuously for $15$ minutes. The acquisition of one entire volume lasted $\sim$30 s with a conventional raster scan pattern. A volume was composed of $156$ B-scans separated $\sim 4.5$ $\mathrm{\mu}$m ($0.7$ mm in total along the $y$-axis). Each B-scan ($zx$ plane) spanned $0.9\times 2$ mm and consisted of $1024\times 512$ pixels (in $z$ and $x$ respectively). Every A-line was repeated two times, before moving to the next $x$ location, to allow the calculation of the correlation between consecutive scans. After the blood assay and after flushing the VoC with EGM-2, an additional OCT scan of the flushed VoC was performed. During postprocessing, each OCT volume was flattened by aligning each A-line to the top boundary of the chip’s channel.

2.2.3 Correlation mapping

Correlation mapping is employed to distinguish between clotted regions and flowing blood. As demonstrated by Tang et al. [22], the autocorrelation of the OCT signal decreases as blood coagulates. We leverage this idea to create a spatial map of the correlation and determine the presence of a thrombus by analyzing the difference in the correlation signal between two consecutive scans. For this method, the intensity from the first scan $I_{t, A}$ is correlated to the intensity of the second scan $I_{t, B}$ at each spatial location $(z,x)$ and time $t$. By evaluating the correlation across an entire B-scan, a correlation map $C_t(z,x)$ is calculated as [22,58]

The correlation window spans $5\times 5$ pixels ($4.4\times 19.5$ $\mathrm{\mu}$m in $zx$) as this offered a reasonable trade-off between resolution and noise. The resulting correlation map ranges between $-1$ and $1$, indicating weak and strong correlations, respectively. While low correlation values indicate red blood cells undergoing rapid motion, strong correlations represent static tissue or thrombi moving at a lower velocity [22]. Commonly, high correlation values $\geq 0.6$ have indicated static tissue that does not undergo any motion [58–60]. However, in the case of forming thrombi, they may present slow motion and weak diffusivity. Therefore, we defined a forming thrombus when the correlation remains $\geq 0.4$ in our correlation maps. The correlation was retrieved for every acquisition, yielding a three-dimensional correlation map. This map was transformed into two-dimensional en-face views ($xy$ plane) by averaging $C_t$ over the height of the VoC (50 $\mathrm{\mu}$m in $z$, Fig. 1(b)). The en-face correlation maps were binarized by thresholding $C_t\geq 0.4$. The thresholded maps at each time interval were utilized to retrieve qualitative and quantitative information from the thrombi. The thrombus size, count, and area were calculated with the function bwboundaries from MATLAB (version 2022a), assuming that the total area of one $xy$ pixel is $17.5$ $\mathrm{\mu}$m$^2$.

3. Results

3.1 Correlation-based thrombus formation

First, we evaluated thrombi formation by analyzing the evolution of the correlation map as a function of time. Figure 2 shows the OCT intensity and correlation map during the blood perfusion assay. En-face and cross-sectional views of the OCT intensity from the VoC at the beginning of the assay are displayed in Fig. 2(a). From these cross-sectional views, it can be observed that the flowing blood produces a homogeneous intensity distribution inside the flow channel, while the boundary of the channel has higher intensity due to the presence of HUVECs and medium change. En-face images were produced by averaging the intensity in the red box in the cross-sections of Fig. 2 within the channel boundaries. The corresponding correlation map in Fig. 2(b) presents low correlation ($C_t\leq 0.4$), which indicates homogeneous flow. Small patches with high correlation $C_t\geq 0.4$ arise at the beginning of the assay, which are attributable to early thrombus formation. As time progresses, Fig. 2(c) exhibits low-intensity regions. These can be related to developing thrombi, where the accumulation of red blood cells attenuates more light than the free-flowing erythrocytes. The correlation map in Fig. 2(d) confirms the development and expansion of thrombi as the correlation value increases over time for the regions where lower flow is expected due to thrombus formation. The two spots indicated by the red arrows show the origin of two thrombi that become visible $\sim 7$ minutes into the assay. In their surrounding areas, the correlation is higher as time passes and the thrombus size increases. In contrast, the blue arrow indicates an area where homogeneous flow is present, and the correlation map has low variation compared to those where thrombi have formed.

 

Fig. 2. OCT intensity and correlation at selected times during the blood perfusion essay. (a) En-face and cross-sectional views of the VoC with flowing blood at the beginning of the blood perfusion assay, the channel boundary is visible in the cross-sectional views. (b) En-face view and cross-sections of the correlation map $C_t$ overlaid with the intensity in (a) derived with Eq. (1) from (a). En-face images in (a) and (b) were produced by averaging the values within the red box indicated in the cross-sections. (c) En-face intensity and (d) correlation as a function of time during the blood perfusion assay. Red arrows indicate the origin of a thrombus. The blue arrow indicates a region without clotting. A video of the entire perfusion assay can be found in Visualization 1.

Download Full Size | PDF

3.2 Thrombus formation validation

After the blood perfusion assay, the flow channel was flushed with the cell culture medium and an additional OCT scan was performed. Figure 3(a) shows the OCT intensity of the flushed VoC, where the final thrombi are visible. The cross-sectional views reveal that thrombi occupy the full height of the channel. In Fig. 3(b), the last correlation map, just before flushing, displays a significant portion of the chip where the correlation is below the defined threshold $C_t\leq 0.4$. Although some thrombi in the correlation map nearly occupy the entire height of the channel (red arrow, Fig. 3(b)), the presence of blood flow (blue arrow, Fig. 3(b)) was maintained and observed throughout the entire experiment. En-face images in Fig. 3(a) and 3(b) reveal a similar location where thrombi are expected from the correlation and are visible after flushing. The slight shift in the $x$-axis in Fig. 3(a) is due to the minor movement of the chip after the tip with blood (Fig. 1(b)) was removed.

 

Fig. 3. Thrombus formation and validation with microscopy images. (a) OCT intensity of the VoC after flushing. (b) Last correlation map before flushing the VoC, exhibiting the presence of thrombi. The red arrow indicates a clotted region inside the channel. As pointed out by the blue arrow, thrombi do not occupy the entire height of the channel, allowing blood to flow in the upper part of the channel. (c) OCT images from the regions of interest I and II in (a) and (b), and respective comparison with images obtained in the microscope consisting of brightfield, fibrin and platelets markers, and the corresponding overlay.

Download Full Size | PDF

Figure 3(c) shows a direct comparison between the OCT signal and microscopy images taken after the blood perfusion assay at two different locations in the microfluidic channel, indicated by box I (solid green line) and II (dashed blue line) in Fig. 3(a) and 3(b). During the blood perfusion assay, the thrombi produce lower intensity and high correlation. In contrast, after flushing, thrombi display higher intensity since scattering arises only at these locations. Because the blood was pre-stained with the fluorescent markers for fibrin (purple) and platelets (green), thrombi were imaged after the blood perfusion assay. The first and second rows of the regions of interest I and II in Fig. 3(c) show a similar shape and size among the correlation map, the OCT flushed signal and the microscopy images. The overlay image shows activated platelet adherence and fibrin threads at the site of the suspected thrombi. A visual comparison between the brightfield image and the flushed OCT channel reveals a similar pattern, differentiated by the lower resolution of OCT compared to the microscope.

3.3 Thrombus growth

Finally, we evaluated the development of the thrombi and the total clotted area over time. Figure 4 presents the evolution of the thrombus formation in the channel over time. After thresholding the correlation maps, a specific color was assigned for a given time, assigning the correlation as a clotting marker over time. Figure 4(a) is the color-encoded representation of the correlation marker during the perfusion assay. At the beginning of the assay, the boundaries of the channel have small thrombi that remain attached to it until the end of the process. The initial thrombi are visible short into the assay ($\sim 5$ min) and highlighted with purple arrows in Fig. 4(a). After $\sim 10$ min, the thrombi have minimally expanded, and the channel allows free blood flow. After $\sim 12$ min, the thrombus indicated by the right purple arrow has grown significantly and continues to expand until the end of the assay ($\sim 15$ min). By the end of the assay, a large thrombus forms around the left purple arrow, and smaller thrombi emerge through the channel.

 

Fig. 4. Thrombus growth and evolution over time. (a) Color-coded location of clotted blood over time. The color represents the time at which a $C_t\geq 0.4$ was registered. The clotted area increases over time as the assay progresses. The purple arrows indicate the presumed origin of two thrombi. (b) Violin plot showing the distribution of thrombi over time within the red box in (a) (excluding the flow channel boundaries). As the assay progresses, bigger clots appear as red blood cell aggregates grow. (c) Total percentage area of the channel with thrombi as a function of time. A video of the thrombus growth can be found in Visualization 2.

Download Full Size | PDF

Figure 4(b) summarizes the dynamic behavior of thrombi formation. The violin plot (obtained from the area inside the red box in Fig. 4(a)) shows that for the majority of the process ($\sim 11$ min), the thrombus count and size remain constant. However, at the later stages of the process, bigger thrombi emerge and expand. Figure 4(c) demonstrates that the total area covered by thrombi inside the flow channel increases over time as a percentage of the flowing area.

4. Discussion

In this work, we demonstrated that OCT can visualize the spatial and temporal changes during thrombi formation without fluorescent markers in flowing blood. A microfluidic VoC device mimicked the interaction between human-whole blood and endothelial cells, enabling localization of the thrombi. The VoC combined with a vis-OCT system allowed simulation and imaging of thrombi formation based on the correlation of the OCT signal. With the correlation between two consecutive B-scans, we observed thrombi formation by thresholding the increase in correlation due to reduced blood flow during thrombus formation. Our observations were validated with fluorescence microscopy images highlighting fibrin and platelets as thrombus markers. The correlation map at the end of the perfusion assay, the intensity after flushing the VoC channel, and the microscopy images showed similarities in location, size, and geometry. Finally, we quantified thrombus formation by measuring the thrombus size and chip thrombus-covered area as a function of time. In the remaining discussion, we review relevant aspects of our results.

4.1 Thrombus formation monitoring

From the OCT data, we could detect thrombi formation in the VoC. As presented in Fig. 2, the OCT signal shows thrombi as a lower-intensity signal, indicating more scattering and absorption by red blood cells. The presence of thrombi is even more notorious when analyzing the correlation signal. We found that if the size of the high correlation area increases over time (as presented in Fig. 4), those spots mark the origin of the thrombi. However, we recognized that the constant blood flow washed out some of the thrombi from the field of view. We attribute this behavior to forming thrombi lacking adherence to the cell layer. The $zx$ cross-section in Fig. 3(a) shows that the thrombi seem connected to the upper and bottom cell layers, but some thrombi do not cover the extent of the chip. This unattached thrombus could not be observed during the blood perfusion assay, due to the constant change in correlation produced by the blood flow, as shown in the $zx$ cross-section in Fig. 3(b). Another limitation of the vis-OCT system that was employed in this study is the low acquisition rate compared to wavelength-swept source OCT (SS-OCT) systems [61,62]. Increasing the sampling rate could reveal thrombus expansion in three dimensions. However, the availability of SS-OCT systems in the visible wavelength range is limited, and higher resolution (in the order of $5$ $\mathrm{\mu}$m) is beneficial to observe thrombus formation [63].

4.2 Correlation mapping as thrombus marker

We noticed that the global correlation of the OCT signal increases over time during the blood perfusion assay. We attribute this behavior to the overall change in the viscoelastic properties of the blood and the expected coagulation over the assay duration [22]. In the flowing blood, thrombi attached to endothelial cells do not undergo the rapid motion of the free-flowing red blood cells, thus increasing the local correlation when compared to the rapid change experienced by moving red blood cells. Therefore, we defined a thrombus as a region where the correlation $C_t\geq 0.4$ in the flowing blood. The correlation threshold for forming thrombi can vary among OCT systems since the correlation depends on the camera line rate, the spatial sampling, and the number of B-scans averaged [58,59]. Additionally, multiple scattering has been shown to affect the decorrelation rate of flowing whole-blood [64]. In our model, we have assumed single scattering behavior inside the relatively small height of the channel ($50$ $\mathrm{\mu}$m) and we assumed that the rate of decorrelation induced by multiple scattering is below the defined threshold. However, multiple scattering effects by blood may contribute to image blurring of the thrombi in the correlation maps, compared to the intensity images of the flushed channels (Fig. 3(a)). A limitation of the method is that we do not measure platelets or fibrin aggregation directly, but rather the changes in flow, which could lead to false positives in the data. A shadow effect can be presented in front of the thrombus in the propagation axis of the beam due to a limited flow which is not due to an aggregate itself [59]. However, looking at the data after flushing, we have captured the final thrombus reasonably well. The shadow effect is negligible for endpoint measurements.

4.3 Validation with microscopy

As presented in Fig. 3(a) and Fig. 3(b), visual inspection of the blood clots show a correspondence in location, shape, and size. However, due to the intermediate flushing step, the exact size and shape of the thrombi before flushing is unknown. Therefore, our correlation-based estimation of the thrombus size and area may be biased compared to the actual thrombus size after flushing. Figure 3(c) exhibits the limitation of conventional OCT intensity to identify blood clots since the contrast is low compared to the flowing blood. The brightfield image and fibrin and platelet markers in Fig. 3(c) have a similar appearance to the intensity of the flushed chip in OCT, even though OCT has a lower resolution. These images reveal a high correspondence between $C_t$ at the end of the process and the location of the blood clots.

4.4 Blood-vessel on a chip device

The VoC was based on microchannels with rectangular geometries, these channels are $300$ $\mathrm{\mu}$m wide and $50$ $\mathrm{\mu}$m high, which does not represent the human artery with respect to size as well as shape. The size of human arteries can range from $25$ to $60$ mm. To overcome the differences in size, the flow rate was meticulously chosen to mimic an arterial shear rate of 1000 s$^{-1}$. The rectangular shape of these VoCs is an abstraction of the actual anatomy of blood vessels in vivo. In the VoC, flowing blood components will experience a lower forward velocity at the boundary of the rectangular geometry compared to a circular channel, leading to higher residence time and exacerbated fibrin deposition. Consequently, we excluded the boundary areas from the analyses of thrombus formation (outside the red box in Fig. 4(a)), as described previously [37], these areas induce artifacts that can affect the data analysis. Cylindrical VoC models [65–67] could be used to mimic the in vivo geometry more closely. The height of these cylindrical models is $\sim 250-300$ $\mathrm{\mu}$m and OCT could unveil the interaction between endothelial cells and flowing blood in arterioles and venules. In our study, HUVECs were used to line the VoC. Although they are a commonly used cell type in blood vessel modeling, these cells can only be obtained from newborns, and not from specific patients, thereby limiting the applicability of the VoC in disease modeling. Therefore, similar research on using OCT to analyze blood clotting in VoCs with human induced pluripotent stem cells-derived endothelial cells (hiPSC-ECs) could determine patient-specific clotting mechanisms and responses.

5. Conclusion

In conclusion, we have used vis-OCT to monitor the spatial and temporal changes in the formation of thrombi in a VoC device. The VoC containing HUVECs simulated the interaction between flowing blood and endothelial cells. The OCT correlation signal provided a marker to observe the evolution of thrombi over time. By thresholding the correlation, we tracked the origin and growth of thrombi. The thrombi detected from the correlation signal were similar in shape and size when comparing the OCT intensity measurements after flushing the chip and the microscopy images. Those microscopy images validated the presence of thrombi. As this pattern is also present in the correlation map, our results suggest the possibility of live image thrombus formation without the necessity of fluorescent markers. Furthermore, we were able to visualize the thrombus formation in width and length and provide a framework for further analysis in height. This dimension could not be determined in real-time using previous fluorescent marker experiments, but only as an end-point measurement using confocal microscopy. Thus, we show a new non-invasive technique to measure thrombus formation in real-time in three dimensions. This technique can now be applied in different, more complex, VoC systems to obtain a greater understanding of thrombus formation and development and with that personalized drug development.