Towards noninvasive white blood cell count

Abstract

Despite rapid advances in diagnostic and imaging technologies, no clinical device for noninvasive monitoring of the immune system is currently available. The standard white blood cell count (WBCC), a key clinical measure for assessing patients’ health, requires drawing blood, which poses inherent risks for secondary infection and anemia in vulnerable patient populations. In addition, the specialized equipment, expertise, and infrastructure are not always available in resource-poor settings. We present a method for noninvasive and label-free WBCC by imaging human oral mucosa using a miniaturized oblique back-illumination microscope (mOBM), together with an analytical pipeline for cell detection and quantification. In a pilot study involving 34 healthy subjects, we validated the ability of our system to detect and quantify circulating leukocytes and compared our image-based WBCCs with standard laboratory measurements. The ability to perform noninvasive WBCC will enable real-time assessment of immune status during infection and inflammation or in response to therapeutic intervention without repetitive blood sampling.

1. Introduction

The white blood cell count (WBCC) is a cornerstone in clinical practice and is widely employed in preventive and emergency medicine to diagnose infection, inflammation, and monitor the response to treatment [1]. Standard blood tests require phlebotomy, which can be challenging in vulnerable patient populations as it poses a risk of infection in immunocompromised patients or anemia in preterm infants [2]. A method for noninvasive measurement of WBCC could alleviate phlebotomy-related complications and enable safe assessment of immune status in vulnerable patient populations. Another key application lies in the preventive monitoring of healthy populations, where normal but elevated leukocyte counts have been linked to disguised systemic inflammation associated with disease development, particularly in elderly and high-risk populations prone to developing cancer, diabetes (type 2), and cardiovascular diseases, among others [3]. Frequent monitoring of immune status with mOBM could enable the early detection and treatment of subclinical conditions.

The development of noninvasive methods for obtaining WBCC has been hindered by the absence of suitable instrumentation that can access patient vasculature, accurately resolve circulating white blood cells (WBCs) in rapidly flowing blood, and record data for a sufficient duration to ensure reliable measurements, as well as lack of analytical tools for objective cell detection and quantification. Commercially available handheld vital microscopes can monitor microvascular perfusion but do not provide a contrast mechanism for delineating individual white blood cells in the circulation [4,5]. Reflectance confocal microscopy [6–8] and spectrally encoded confocal microscopy/flow cytometry [9] are capable of high-resolution cellular imaging in vivo, but the presence of speckles compromises image quality. Moreover, strong backscattering from both red and white blood cells results in poor contrast between the two cell types when intermixed in the flowing bloodstream [8,9]. Using adaptive optics, it is now possible to obtain exquisite images of local immune cell dynamics in the retina in vivo [10,11], but whether the retina can serve as a window into the systemic immune system is yet to be demonstrated. Nonlinear optical techniques, such as third-harmonic generation [12] and two-photon-induced autofluorescence microscopy [13], have been explored, but their complexity and high instrumentation costs pose a formidable barrier to clinical translation.

Herein, we present a method for noninvasive and label-free WBCC by imaging the human oral mucosa using a miniaturized oblique back-illumination microscope (mOBM) that provides phase-gradient contrast in thick tissues [14,15], enabling individual blood cells to be resolved when capturing images with sufficiently short exposure times (0.5 ms) and high frame rates (up to 300 fps). In addition, we present an analytical pipeline for cell detection and quantification and compare our image-based WBCC with standard laboratory analysis in 34 healthy volunteers.

2. Materials and methods

2.1. Miniaturized oblique back-illumination microscope (mOBM)

The instrument illustrated in Fig. 2(a) is an improved version of the recently reported device [16]. Its optical layout was designed using the Zemax Optical Studio software (Fig. S1(a) in Supplement 1 ^{(3.3MB, pdf)} ), achieving a compact system (Fig. S1(b) in Supplement 1 ^{(3.3MB, pdf)} ) with only four components and weighing just 0.130 kg.

Fig. 2.

Imaging the microvasculature and leukocytes using the miniaturized oblique back-illumination microscope (mOBM).(a) The miniaturized oblique back-illumination microscope (mOBM). (b) Exposed vasculature bed of the lower lip. (c) Improved visualization of the vasculature in the lip compared to the skin tissue. (d) Examples of leukocytes in the microvasculature imaged with our mOBM. Large cells (≥10 µm), most likely neutrophils, are indicated with red arrowheads (i-iii, vi). Small leukocytes (≤10 µm), most likely lymphocytes, are indicated with green arrowheads (iv-vi). (e) Individual RBCs (blue arrowheads) in the capillary flow. (f) A sequence of images of a rapidly flowing leukocyte (red arrowheads) acquired at 250fps. (g) Space-time diagram reconstructed using the pixel intensity along flow direction as a function of time across 200 consecutive images ( Supplement 1 ^{(3.3MB, pdf)} ); I: Initial (pre-processing) F: Final (after processing). (h) Example of blood flow velocity and heartbeat measurements as a function of time. Heartbeat was derived from the intensity fluctuation within an ROI outside the blood vessel due to slight tissue deformation with each heartbeat.

The system was designed with a high numerical aperture (NA = 0.75, water immersion) aberration-corrected imaging GRIN lens assembly (Grintech, GT-MO-080-032-ACR-VISNIR-08-20). This design provides a resolution of ∼0.9 ± 0.1 µm in xy, measured using 0.5 µm fluorescent microspheres. The magnification of the GRIN lens was 2.2, and the total magnification was 13.45 after the relay optics, yielding approximately 3.9 pixels/µm on the CMOS camera without binning and 1.95 pixels/µm with binning (2 × 2 pixels). Given the selected magnification, the diameter of the imaging GRIN lens, and the 6.6mm × 4.1 mm sensor size, the resulting field of view (FOV) was ∼360 × 300 µm. The high-speed CMOS sensor (Basler, daA1920-160um) capable of image acquisition up to 1000 frames per second was used to enable efficient detection of rapidly moving WBCs across multiple frames. To generate oblique back-illumination, we used an epi-illumination configuration wherein the optical fiber (multimode, 1 mm core diameter) was positioned next to the imaging GRIN lens assembly (Fig. S1(c) in Supplement 1 ^{(3.3MB, pdf)} ). This technique provides phase-gradient contrast (PGC), which reveals fine morphological details, such as cell border and intracellular granules of blood cells (Fig. 1(d)). In addition, we use a 565 nm light emitting diode (LED, Thorlabs, Solis-565D) that provides absorption contrast between red and white blood cells. The output of the LED is coupled into the optical fiber, and the power at the output end of the optical fiber is typically ∼30 mW, below the maximum permitted exposure limit of 34 mW.

Fig. 1.

Analytical pipeline for WBCC quantification. (a). Raw image representing a video acquired by mOBM. (b). Co-registration ensured the alignment of image features across video frames. (c). Flat-field correction ensured uniform illumination across each video frame. Brightness and contrast adjustments were also performed at this step (d). The Area of Analysis (AOA) was bound by the automatically-generated vessel luminal mask (yellow), which identified the locations of the vessel walls, and by a user-defined mask (teal), which informed the software of the endpoints of the AOA. (e). A coordinate system was created within the AOA. Its x’-axis was the flow axis (left), and x’-gridlines spanned the vessel’s luminal width (right). The vessel luminal diameters were the width values at each pixel along x’ (purple arrow). (f). Blood flow velocity was determined using space-time diagrams (left), which showed the motion of dark and bright objects moving along the flow axis (x’) over time (t). The slope of the stripes were inversely proportional to the flow velocity. Blood flow volume was calculated by multiplying the flow velocity values to the luminal cross-sectional area and elapsed imaging time (right). (g). The WBC count was a cell count window’s (left) peak count of a temporal trace of mean pixel values (right). (h). The ROI’s WBCC values over time was evaluated for stability. Only those that demonstrated minimal changes (<5%) over a significant sampling volume (≥20nL) were accepted. Scale bar = 20 µm.

A key element of the mOBM is the imaging tip [16], designed with four channels (Fig. S1(c) in Supplement 1 ^{(3.3MB, pdf)} ): a central channel housing the miniaturized GRIN objective lens, an illumination channel holding the tip of the optical fiber, an irrigation channel to prevent the oral mucosa tissue from drying and to maintain the immersion medium (water) in front of the GRIN lens, and a ring-shaped vacuum cavity to stabilize the imaging tissue when needed. The working distance (WD) of the imaging GRIN lens assembly (Fig. S1(a) in Supplement 1 ^{(3.3MB, pdf)} ) was adjusted using motorized camera displacement along the optical axis. The mOBM’s mechanical housing, imaging tip, and oral mucosa apparatus (Fig. 2(c)) were designed using SolidWorks software and printed using a 3D laser printer (Formlabs 3B). All parts in contact with human tissue were printed using biocompatible materials (Formlabs, RS-F2-BMCL-01, and RS-F2-BMBL-01).

2.2. Oral mucosa apparatus for lip stabilization

To image the microvasculature in the oral mucosa, we gently exposed the lower lip using a universal oral mucosa apparatus developed in our laboratory [16]. The mechanical contours of the apparatus were tailored to conform to lips’ morphology and minimize the high-pressure points that could potentially disrupt blood flow.

2.3. Imaging of healthy participants

Imaging and data collection started after the participants (healthy volunteers) signed the IRB-approved consent form (MGH Protocol #:2021P003047). Participants were seated in front of the imaging system, with the chair and chin holder adjusted accordingly. The lower lip was then gently unrolled using the oral mucosa apparatus, followed by an imaging session that typically lasted 30-50 min. During imaging, the operator used a motorized XYZ actuator (mounted with mOBM) with a micrometer resolution to navigate through the exposed lip tissue (Fig. 2(b)) to identify and record the vessels of interest.

2.4. Image acquisition parameters

Pylon camera software (Basler) was used for image acquisition. The acquisition frame rate was 200-300 fps, and binning (2 × 2) was often used to increase the signal-to-noise ratio (SNR), reduce the exposure time (0.5 ms), and decrease the size of recorded video files (.avi).

2.5. Measurements of vessel luminal diameter, blood flow velocity, and WBC count

Video frames acquired by the mOBM (Fig. 1(a)) were co-registered to reduce residual motion (Fig. 1(b)), and flat-field correction was applied to uniformize background illumination (Fig. 1(c)). The videos were then fed into a custom software for measurements of vessel luminal diameter, blood flow velocity, and WBC count. The workflow is summarised in Fig. 1. Further details are provided in Fig. S2-S6 of Supplement 1 ^{(3.3MB, pdf)} .

The custom software first identified an area of analysis (AOA), which was bound by an automatically generated vessel luminal mask and a user-defined mask (Fig. 1(d), Fig. S2 in Supplement 1 ^{(3.3MB, pdf)} ). The vessel luminal mask was characterized by pixels with frequent and significant fluctuations in video-frame-to-video-frame pixel values. The user-defined mask informed the software of the area’s endpoints. No further user input was required after this step.

The software subsequently created a coordinate system within the AOA (Fig. 1(e), Fig. S2 in Supplement 1 ^{(3.3MB, pdf)} ). The blood flow axis was defined as the x’-axis. The width of the coordinate system spanned the width of the vessel lumen, previously identified by the vessel luminal mask. The width values measured at each pixel along x’ were recorded as the vessel luminal diameters. The length of the coordinate system was bound by the AOA and optimized for velocity measurements. The extent of the coordinate system was the region-of-interest (ROI).

The software then measured the blood flow velocity using space-time diagrams, which were dimensional re-arrangements of the input video such that the intensity variation along a line in the flow direction (x’) was plotted as a function of time (t). (Figure 1(f), Fig. S3 in Supplement 1 ^{(3.3MB, pdf)} ). The stripes in the diagrams traced the motion of their corresponding objects (dark = red blood cells (RBCs), bright = WBCs or plasma surrounding RBC), hence, their slopes were inversely proportional to the objects’ velocity and sensitive to the objects’ space- and time-dependent velocity changes. Blood flow velocity as a function of time was determined from the statistics of the weighted slope values of the RBC stripes. Blood flow volume was the multiple of the velocity values, the luminal cross-section area calculated from the vessel luminal diameters, and elapsed imaging time.

Next, the software performed WBCC measurement (Fig. 1(g), Fig. S4 and S5 in Supplement 1 ^{(3.3MB, pdf)} ) by counting the peaks in a background-corrected temporal trace of mean pixel values within a cell count window, the size and position of which were determined automatically by mapping the local contrast and vessel lumen diameters within the ROI. The threshold for peak finding was defined as 3.5× the scaled mean absolute deviation of the residual noise after background correction. WBCC was the WBC count divided by the blood flow volume obtained in the previous step.

Finally, the software performed a stability check of ROI’s cumulative WBCC value over time (Fig. 1(h), Fig. S6 in Supplement 1 ^{(3.3MB, pdf)} ). This step ensured the local blood flow had been sufficiently sampled for reporting. The WBCC stability criterion was fulfilled by the ROI if the volume-binned cumulative WBCC values in the last ≥20nL of the ROI’s imaged flow volume fell within 5% of the final binned cumulative WBCC value.

2.6. Boltzmann fit for WBCC correction

The efficiency of WBC detection using mOBM is optimal when cells are in focus and are not masked by surrounding RBCs. In capillaries and small vessels, WBCs occupy the entire lumen, unobscured by RBCs (Fig. 2(d), i-iv), and nearly all WBCs within the sampled blood volume were detected. However, as the vessel diameter increases, the lumen cross-section can accommodate multiple cells, and WBCs may be outside the imaging plane or masked by RBCs that travel more superficially through the blood vessel (i.e., between the WBC and the microscope). In these situations, WBCs can be missed, leading to an underestimation of WBCC (Fig. S7 in Supplement 1 ^{(3.3MB, pdf)} ). The larger the diameter of the blood vessel, the higher the chance that WBCs will go undetected and the lower the WBCC measurement compared to the reference value (Fig. S7(a) in Supplement 1 ^{(3.3MB, pdf)} ).

To model the underestimation of WBCC detection with mOBM, a binary logistic regression model was represented by a Boltzmann function. This model is appropriate for our case, in which binary cell detection (detected or not detected) is a function of the blood vessel diameter. For small vessels, the Boltzmann function approaches 1 (Fig. S7(a) in Supplement 1 ^{(3.3MB, pdf)} ) owing to the efficient WBC detection. However, the fit curve gradually decreased as the number of missed WBCCs increased in larger vessels. The fitted Boltzmann function was used to correct the corresponding raw WBCC measurements, accounting for leukocytes that were not detected within the sampled blood volume.

3. Results

We have previously reported a miniaturized phase-contrast microscope [16] capable of in vivo monitoring of leukocyte-endothelial interactions, such as rolling and adhesion, which are slow processes that can be monitored at 30-50 fps. In this work, we investigated the ability of our mOBM system (Materials and Methods) to detect and quantify rapidly flowing leukocytes in the human microcirculation (Fig. 2(a)-(c), and Fig. S1 in Supplement 1 ^{(3.3MB, pdf)} ). We chose the lower lip as the imaging site because of its favorable characteristics [9,16], including the large perfusion bed (Fig. 2(b)-(c)), thin stratum corneum layer, low pigmentation compatible with individuals with dark skin color, comfort for the subjects, and ease of exposing and stabilizing tissue while minimizing motion artifacts during imaging (Materials and Methods). For imaging, we employed the oblique back-illumination (OBM) approach that generates phase-gradient contrast in thick tissue [14,15,17], producing clear delineation of cell borders, which is essential for detecting and distinguishing individual WBCs from plasma gaps (Fig. 2(d), i-v, Supplement 1 ^{(3.3MB, pdf)} ( Visualization 1 ^{(80.2MB, avi)} and Visualization 2 ^{(80.2MB, avi)} ). WBCs that travel together as doublets or triplets can also be visualized (Fig. 2(d), vi). Additional absorption contrast is generated by the 565 nm illumination wavelength, which is strongly absorbed by RBCs [15], rendering them darker and readily distinguishable from the much brighter WBCs (Fig. 2(d), (f)). Other key features of the mOBM are the short exposure time (≤0.5 ms) that reduces motion blur and the high acquisition frame rates (up to 300 fps) that enable frame-to-frame detection and tracking of rapidly moving cells in the bloodstream (Fig. 2(f)).

To measure the blood flow velocity, our image processing pipeline (Materials and Methods and Supplement 1 ^{(3.3MB, pdf)} ) plots the pixel intensity along the flow direction as a function of time to generate space-time diagrams, where the slopes are the inverse of the flow velocity values (Fig. 2(g)-(h)). In a subset of blood vessels, presumably arterioles, the velocity fluctuated in sync with the heartbeat (Fig. 2(h)). The blood volume rate can then be obtained by multiplying the velocity by the cross-sectional area of the vessel. WBCC is defined as the number of WBCs per unit volume.

To count WBCs, our image-processing pipeline automatically defines a cell-count window within the ROI (Fig. 3(a)). Because of the differential absorption between WBCs and RBCs, the brightness of the pixels within the window increases when a WBC is present. Figure 3(b) shows an example of a temporal trace obtained by plotting the mean pixel intensity within the window as a function of time after background correction. The peak positions indicate the video frames in which the leukocytes appear in the window. The pipeline then automatically counts the peaks whose intensity surpasses a moving threshold (Fig. 3(b), dotted grey line). To assess the accuracy of automated peak detection, we compared the results with manual cell counts and found good agreement between the measurements (Fig. 3(c), ratio = 0.96 ± 0.11). Note that manual cell counting is only possible for blood vessel diameters smaller than ∼ 22 µm. In larger vessels, some leukocytes may be out of focus in the flow stream, and leukocytes in deeper parts of the blood vessel can be obscured by RBCs at shallower depths, making manual identification progressively more difficult as vessel size increases.

Fig. 3.

Automated leukocyte detection, quantification, and stabilization of WBCC measurements.(a)-(b) Method for counting circulating leukocytes. The mean of pixel values in the cell count window marked by the red dotted line in (a) is displayed over time, as shown in (b). The peaks that exceed the time-dependent threshold value (dotted gray line) are counted by the automated pipeline as WBCs ( Supplement 1 ^{(3.3MB, pdf)} ). The purple stars indicate the peaks identified by the automated pipeline, and the green checkmarks indicate the manually identified WBCs in the corresponding video. (c) The ratios of automatic over manual cell counts of the same video ( Supplement 1 ^{(3.3MB, pdf)} ). (d) Example of cumulative WBCC as a function of acquisition time for a single vessel, which fulfilled our stability criterion (<5% fluctuation) in ∼101 seconds. In total, 81 vessels (from 29 subjects) fulfilled the stabilization criteria. (e) The distribution of stabilization time. Each data point represents a single vessel. (f) The ratio between the corrected WBCC obtained by our imaging method divided by the reference value obtained by drawing blood. Each point is an individual measurement from a single blood vessel. The 95% confidence interval (CI) of a linear fit is indicated with the blue color. The green arrows indicate the narrowest CI area, suggesting the optimal vessel diameter to be centered at ∼17 µm. (g) Ratio variations when averaging measurements from multiple vessels from the same subject. Each data point is from a single subject (N = 29). Subjects with stabilized counts in only a single blood vessel are represented by the gray bar, while subjects with stabilized counts in 2 or 3-5 vessels are represented by blue, and green bars, respectively. The error bars represent the standard deviation (SD). The mean and SD are indicated at the top of the box plots.

Next, we used mOBM to image the microvasculature in a cohort of 34 healthy volunteers and compared our image-based WBCC with existing clinical standards obtained by drawing blood from the same subjects after imaging. We analyzed 128 blood vessels to investigate the minimum blood volume (and hence the acquisition time or the length of video) needed to obtain a reliable WBCC measurement. As shown in Fig. 3(d), the cumulative WBCC initially fluctuated but eventually stabilized as the imaging time increased. Of the 128 vessels analyzed, 81 from 29 different subjects fulfilled the stability criteria (see Materials and Methods). The acquisition time needed for stabilized WBCC measurements was 3 ± 2 min (Fig. 3(e)), corresponding to a sampling volume of 63 ± 61 nL, depending on the vessel size and flow velocity.

We evaluated the ratio R between the WBCC obtained from mOBM and the standard laboratory test after blood draw (R = 1 if the two measurements were in perfect agreement). As shown in Fig. S7 in Supplement 1 ^{(3.3MB, pdf)} , R obtained from individual vessels depends on vessel size, decreasing with increasing vessel diameter, as a greater fraction of WBCs is either out of focus or masked by more superficial RBCs. We applied a correction factor to account for undercounting at larger vessel diameters (see Materials and Methods). The result after the correction is shown in Fig. 3(f). Interestingly, the confidence interval (CI) of the linear fit presented in Fig. 3(f) is narrower for 15-19 µm diameter vessels, suggesting that this size range may be preferable for the most consistent WBCC measurements. Coincidentally, they are also the most abundant vessel sizes in the superficial oral mucosa.

Finally, we report the WBCC obtained from mOBM as the average count from multiple blood vessels within the same subject. Averaging measurements from multiple vessels should naturally lead to the reference WBCC value since phlebotomy samples blood from large veins that drain numerous microvessels. As shown in Fig. 3(g), the ratio starts to converge and remains close to 1 as the number of averaged vessels increases. Importantly, the mean absolute percentage error (MAPE) of WBCC measurement decreases from 74% to 30% when imaging 3-5 vessels compared to a single vessel, leading to more precise measurements that align closer to the reference WBCC.

4. Discussion

We present a miniaturized microscope system for non-invasive and label-free monitoring of systemic white blood cell count (WBCC) that is a hallmark of inflammation and a key clinical indicator of patient health.

To validate our system, we imaged 34 healthy volunteers, demonstrating the ability to acquire high-quality videos across a wide spectrum of skin colors, ages, and sexes (Fig. S8 in Supplement 1 ^{(3.3MB, pdf)} ). By acquiring temporal recordings from various vessels of the oral microvasculature and assessing the WBCC over time, we showed that, after initial fluctuations, the WBCC measurement stabilized in most of the vessels (81 out of 128, or ∼63%). Obtaining WBCC from small vessels requires long acquisition times (3-5 min per vessel) because of the small sample volume paired with the stochastic arrival of WBCs. The resulting noise within incremental volume packets (Fig. S6 in Supplement 1 ^{(3.3MB, pdf)} ) becomes less significant as the total sampled volume increases. However, for reasons that are still unclear, the WBCC in some vessels continues to fluctuate and does not stabilize. We exclude these vessels by imposing the stabilization criteria to ensure consistent and reproducible measurements that accurately reflect the true WBCC in specific vessels rather than transient fluctuations. In addition, we ensure steady blood flow during image acquisition by minimizing the pressure applied to the tissue by the imaging tip, and we avoid looped, branching and inter-crossing vessels, which are less suitable for post-processing analysis. The detection of WBCs in large vessels (>22 µm) using mOBM also presents significant challenges due to poor cell border delineation. Using data from 81 stabilized vessels, we established the first benchmarks for stabilized image-based WBCC measurements, indicating an average acquisition time of 3 ± 2 min, corresponding to a sampling volume of 63 ± 61 nL/vessel. The ability to perform extended temporal recordings is crucial for reliable image-based WBCC measurements and represents a key distinction between our mOBM system and existing commercially available handheld microscopes [4,5], which are often limited to short video acquisitions owing to severe motion and pressure artifacts.

The analysis of stabilized WBCCs showed that detection efficiency declined with increasing vessel diameter (Fig. S7(a) in Supplement 1 ^{(3.3MB, pdf)} ). We empirically derived a vessel-size-dependent correction factor and compared the corrected WBCC measurements with clinical reference values (Fig. 3(f)). The results indicate that WBCC measurements from vessels with ∼15-19 µm diameters align closest to the clinical reference value (Fig. 3(f), green arrowheads). Therefore, our instrument targets vessels within this size range to achieve optimal image-based WBCC. Moreover, arterioles are preferable to venules, whose flow is less stable and is more likely to be interrupted by rolling leukocytes. Arterioles are also preferable to smaller capillaries [18], whose flow is even less robust, with the frequent presence of plasma gaps, which can lead to leukocyte-detection artifacts.

To further improve the accuracy of the image-based WBCCs, we averaged the results from multiple vessels from the same participant (Fig. 3(g)). This approach significantly reduces the mean absolute percentage error (MAPE) from 74% to 30% when averaging measurements from 3-5 vessels, compared with using single-vessel data. This important finding suggests that averaging multiple vessels provides a more accurate assessment of systemic immune status, accounting for uneven leukocyte distribution in the microvasculature, which is determined by the branching order of individual vessels in the vascular tree [19]. In addition, factors such as blood flow velocity and local hematocrit can orchestrate leukocyte margination [20,21], which can impact their partitioning after bifurcation and the overall distribution across the vasculature network. In future implementations, we plan to increase the FOV by reducing the magnification (by a factor of ∼2) while maintaining the resolution (using the sensor without binning) to facilitate navigation and vessel selection within the lip’s microvascular network.

The method described here represents an important advancement toward noninvasive measurement of WBCC in humans, with the potential for substantial clinical impact, such as monitoring critically ill patients in ICUs [22], immunosuppressed cancer patients undergoing chemotherapy [23], and preterm infants [2], for whom the traditional blood draw approach poses a high risk of secondary infections, anemia, and pain. The noninvasive aspect of mOBM also enables continuous monitoring of the immune system, which is critical for timely clinical interventions in cases such as sepsis, wherein every hour of delayed treatment reduces survival chances by 9% [24]. Another key application of mOBM is in preventive healthcare for the early detection and treatment of diseases in the overwise-considered healthy adult population. Indeed, WBC counts at the high end of the normal range have been linked to chronic systemic inflammation and to subclinical disease progression, especially among elderly and people with a higher risk of developing cardiovascular diseases, autoimmune diseases, cancer, type 2 diabetes, and depression among others [3]. Traditionally, monitoring such subclinical states requires repetitive blood draws throughout the day to capture fluctuations in leukocyte levels, a process that is problematic in routine healthcare due to the invasiveness, risk of infection, and potential for anemia, particularly in at-risk groups. Our method can bridge this gap and enable frequent WBCC measurements for determining the immunological setpoint [25] as well as timely diagnosis and management of subclinical and developing diseases. Finally, the low cost and portability of the mOBM system make it particularly valuable in resource-poor settings (e.g., regions that lack infrastructure, such as specialized equipment or advanced healthcare facilities), allowing for a simple noninvasive assessment of immune status in the most difficult circumstances.

5. Conclusion

The miniaturized oblique back-illumination microscope (mOBM) we developed offers a new approach for noninvasive and continuous assessment of systemic white blood cell counts in humans. Our methods address the longstanding limitations of existing systems, offering stabilized in vivo measurements, vessel-specific corrections, and multivessel averaging necessary for achieving reliable WBCCs. These advancements make mOBM particularly promising for critical clinical applications, for safe monitoring of the immune status in ICU patients and preterm infants, as well as for preventive monitoring in high-risk populations. Ongoing work aims to develop a large field-of-view imaging system that will record multiple vessels in a single video acquisition. This will improve the accuracy of mOBM and decrease the imaging time per patient, which is currently between 30-50 minutes. Another key improvement involves the development of an automated machine-learning pipeline to enhance cell detection, accelerate image processing, and classify different leukocyte subtypes (e.g., neutrophils vs. lymphocytes) based on cell size and granularity. Further miniaturization of mOBM will accelerate its integration into clinical practice and facilitate its use in neonates, ICU patients, and resource-poor settings.

Funding

U.S. Department of Defense 10.13039/100000005 ( FA9550-20-1-0063, FA9550-23-1-0656); Massachusetts General Hospital 10.13039/100005294 ( FMD Clinical Award); SPIE 10.13039/100007347 ( Franz Hillenkamp Fellowship).

Disclosures

A.B., J. W., and C.L. have filed a US patent application “System for stabilized noninvasive imaging of microvasculature in the oral mucosa”.

Data availability

The data supporting this study's findings are available from the corresponding author upon reasonable request.

Supplemental document

See Supplement 1 ^{(3.3MB, pdf)} for supporting content.