Acoustofluidic-based microscopic examination for automated and point-of-care urinalysis

Abstract

Urinalysis is a heavily used diagnostic test in clinical laboratories; however, it is chronically held back by urine sediment microscopic examination. Current instruments are bulky and expensive to be widely adopted, making microscopic examination a procedure that still relies on manual operations and requires large time and labor costs. To improve the efficacy and automation of urinalysis, this study develops an acoustofluidic-based microscopic examination system. The system utilizes the combination of acoustofluidic manipulation and a passive hydrodynamic mechanism, and thus achieves a high throughput (1000 μL min⁻¹) and a high concentration factor (95.2 ± 2.1 fold) simultaneously, fulfilling the demands for urine examination. The concentrated urine sample is automatically dispensed into a hemocytometer chamber and the images are then analyzed using a machine learning algorithm. The whole process is completed within 3 minutes with detection accuracies of erythrocytes and leukocytes of 94.6 ± 3.5% and 95.1 ± 1.8%, respectively. The examination outcome of urine samples from 50 volunteers by this device shows a correlation coefficient of 0.96 compared to manual microscopic examination. Our system offers a promising tool for automated urine microscopic examination, thus it has potential to save a large amount of time and labor in clinical laboratories, as well as to promote point-of-care urine testing applications in and beyond hospitals.

---

Introduction

Urinalysis is one of the most requested clinical laboratory tests for monitoring nephrological and urological conditions, screening liver and kidney diseases, detecting urinary tract infections, and so on.^1–5 A complete urinalysis test involves three parts: observation of physical properties, analysis of chemical compositions and microscopic examination of the urine sediment.^6–9 Among them, physical observation is relatively rapid and automated instruments (such as a strip reader) or point-of-care devices for chemical analysis are well-established, thus the first two parts can be conducted easily and efficiently.^10–12 Therefore the microscopic examination—quantitative detection of substances such as cells, casts, crystals and bacteria in the urine sediment—becomes the bottleneck obstacle that hinders the efficacy and automation of urinalysis.^13,14

In recent years, urine sediment analysis tools such as automated microscopy analyzers and urine flow cytometers have been developed.^15–20 Microscopy analyzers can quantify cells in uncentrifuged urine using laminar flow of the specimen through the lens of a charged coupling device (CCD) camera.^15–17 And urine flow cytometers can detect cytometric characteristics of individual cells through measurement of differential light scattering and fluorescence emitted by illuminated cells.^18–20 Though huge efforts have been made, these tools still depend on complicated fluidic and optical systems, and therefore, they are expensive and bulky which cannot satisfy the widespread medical applications and point-of-care diagnosis. As a result, urine sediment microscopic examination remains one of the few manual procedures that are still widely performed in clinical laboratories, making urinalysis labor-intensive and time-consuming with a large inter-observer variation.²¹

In order to improve the automation and efficacy of urinalysis, a compact and low-cost tool for urine microscopic examination needs to be developed replacing the bulky and complicated systems. However, there are two major technical challenges that should be resolved. First, the concentration of cells in urine is very low (usually <10⁴ cells per ml), ∼50 fold lower than the standard of microscopic examination.^15,16 Therefore, to eliminate the burden on downstream detection, it is crucial to develop an automated pre-treatment method with both high throughput and a high concentration factor. Second, the transfer of the concentrated sample to a hemocytometer and the observation using a microscope are currently conducted manually. Therefore, to achieve the “sample-in-answer-out” goal, it is necessary to integrate pre-treatment, micro-dispensing and image acquisition functions into a single device, followed by automated image recognition algorithms.

Microfluidic technology has the ability to manipulate fluid and particles precisely, thus it has the potential to resolve the technical challenges in automated urine microscopic examination.^22–26 Various types of microfluidic-based techniques for sample enriching have been reported using both passive and active methods.^27–34 For example, Xiang et al.³¹ used inertial microfluidics to conduct high-throughput concentration of tumor cells with a concentration factor of 40 for a single run. However, to achieve higher fold concentration, serial cascading is needed, causing inconvenience for automation. Feng et al.³³ reported a deterministic lateral displacement structure that can achieve 50-fold concentration of particles but the throughput is constrained (110 μL min⁻¹). In addition, passive methods typically rely on complicated and subtle geometries of microchannels, making it challenging to handle abnormal urine samples that contain substances with larger sizes, i.e. casts and epithelial cells. Compared with passive methods, active methods use external forces to concentrate cells in the fluid, thus the geometry of microchannels is able to allow substances with various sizes. However, the throughput and concentration factor of active methods usually cannot fulfill the needs of urine microscopic examination. The throughput is constrained (<500 μL min⁻¹) since the retention time for each particle in the external field is limited at higher flow rates.^35–40 On the other hand, due to the addition of sheath fluid that dilutes the sample, the concentration factor is also limited (3 to 10-fold).^41–45 To improve the performance, researchers have tried many methods such as using serial cascading or applying a higher external force.^46–48 For example, Jakobsson et al.⁴⁶ used a recirculating fluid system to achieve thousand-fold concentration; however, the multiple runs posed a challenge for automation. Jonai et al.⁴⁷ used a two-dimensional acoustic focusing method to obtain a high concentration factor; however, the throughput issue was still not resolved (300 μL min⁻¹).

As a result, it is challenging for current microfluidic devices to conduct a urine pre-treatment process due to their low throughput, low concentration factor or other issues. Moreover, samples processed by current microfluidic chips need to be loaded to glass slides manually. Lacking an integrated system with both pre-treatment and sample dispensing modules prevents the automation of downstream analysis.

Herein, to resolve the obstacle that hinders the automation of urinalysis, we develop an acoustofluidic-based microscopic examination system for the concentration and analysis of the urine sediment. We combine acoustofluidic manipulation and a passive hydrodynamic mechanism to achieve high throughput and a high concentration factor simultaneously. The chip consists of three paralleled branch channels and each branch can achieve 4.5 to 4.8-fold concentration. The flow rate of particles dropped down dramatically after flowing through each branch, thus there will be longer travel time for the particles within the acoustic field, which can improve the throughput. As a result, the urine sample is processed with a high throughput (1000 μL min⁻¹) and the cells are enriched by 95.2 ± 2.1 fold. We design an interconnector to integrate a pre-treatment module and a hemocytometer-like microchamber, and therefore, the concentrated sample is automatically dispensed and the cells are distributed in the chamber on the glass slide due to the capillary effect. The images of the enriched sample are acquired under a microscope and segmented via a machine learning algorithm. Our acoustofluidic-based system can perform urine microscopic examination in 3 minutes with detection accuracies of erythrocytes and leukocytes of 94.6 ± 3.5% and 95.1 ± 1.8%, respectively. We test urine samples from 50 patients and the correlation coefficient between our system and the conventional manual microscopic examination is above 0.96. In addition, our system is able to capture epithelial cells, casts, crystals and other urinalysis biomarkers that can be indicators for various diseases. This acoustofluidic-based urine microscopic examination system paves the way for automated urinalysis, and therefore, it is of great potential to boost the efficacy and save significant labor in clinical laboratories. In addition, this acoustofluidic-based urinalysis system is compact and low-cost, and thus, will be beneficial for point-of-care applications.

Results and discussion

Concept of acoustofluidic-based automated urine microscopic examination

A comparison of conventional and the acoustofluidic-based urine microscopic examination procedures is shown in Fig. 1. The conventional procedure typically contains multiple steps including centrifugation, pipetting, slide preparation and manual examination. The whole process is labor-intensive and time-consuming with a large inter-observer variation. In contrast, the acoustofluidic-based urine microscopic examination system developed by us can achieve “sample-in-answer-out” analysis within 3 minutes automatically. The system takes 2 mL urine samples into the integrated microchip where sample pre-treatment and dispensing are performed, getting ready for machine learning-based image analysis. The results are automatically generated via programmed algorithms. The whole process is in one step without humans involved.

Fig. 1 Schematic diagram of human urine microscopic examination. The conventional urine microscopic examination in clinical laboratories is performed manually, requires multiple steps that are time consuming (≈15 minutes) and is labor-intensive. In contrast, the proposed acoustofluidic-based system integrates all steps into a microchip where the urine sample is automatically processed via the combination of acoustic and hydrodynamic mechanisms. Assisted by some accessories and a machine learning algorithm, automated urine microscopic examination is conducted in 3 minutes.

The heart of our system is an acoustofluidic microchip shown in Fig. 2. An adjustable holder is used to install the microchip (Fig. 2A). A temperature sensor and a cooling fan are used to detect the temperature of the device and take away the additional heat generated by the piezoelectric transducer. We tested the temperature change of the microchip using a thermal sensor, and the data are shown in Fig. S1 (ESI†). The temperature can be controlled at 37 °C using the fan and the heat dissipation of the holder. The microchip is mainly composed of two modules: the pre-treatment module and the micro-dispensing module (Fig. 2B). The first one, the pre-treatment module, is used for high-fold concentration of urine samples. The sediment in the native urine sample such as cells, casts and crystals can be concentrated under the combined effect of acoustic and hydrodynamic forces. Specifically, the chip has a trident-like channel network that is composed of a long straight channel in the middle and three symmetrically parallel branch channels. The transducer attached to the bottom of the chip generates half-wavelength standing acoustic waves in the middle channel to form a node at the center line. Thus, the acoustic radiation force pushes particles with positive acoustic contrast factors to move towards the wave node, enriching the concentration of the particles. However, solely using the flow splitting of a specific outlet system can hardly achieve a high concentration factor stably.^34–36 Therefore, to achieve both a high concentration factor (the ratio of cell concentration after and before processing) and throughput (the volume of the urine sample processed in one minute), three serial separation junctions are used. When the suspension arrives at each junction, it splits to three branches, and a bulk of fluid flows to side branches symmetrically. The hydrodynamic effect dominates the movement of particles at each junction and they diffuse to the side direction of the middle channel along with the streamline. But they remain in the middle channel because they are pre-aligned at the center of the channel by the acoustic radiation force. With most of the liquid flowing to the side branches, the concentration of particles in the middle channel increases significantly. Meanwhile, the flow rate of the enriched suspension in the middle channel decreases dramatically after passing through each junction, which ensures enough time for cells to relocate to the middle of the channel driven by the acoustic radiation force so that they will not flow to the branch channel in the next junction. After repeating the process several times, the sample can be concentrated many fold even under a high throughput. The second module of the microchip—the micro dispensing module—is designed to transfer the concentrated sample to a semi-closed chamber similar to a hemocytometer. The height of the chamber is 80 μm and the volume is 18 μL. After that, the processed urine sample is imaged under a microscope automatically.

Fig. 2 The microchip used in the acoustofluidic-based system for urine processing. A) Photograph of the microchip installed in a holder with a piezoelectric transducer, a temperature sensor and a cooling fan mounted. B) Schematic diagram illustrating the working principle of the acoustofluidic-based microchip. Bioparticles in urine are concentrated by acoustic waves and the hydrodynamic mechanism in the trident-like channels, and the enriched sample is flowed to the micro-dispensing module for downstream analysis.

Design and optimization of the acoustofluidic microchip

An appropriate size and structure would improve the total concentration factor and throughput of the chip. Therefore, the geometric dimensions of the acoustofluidic microchip were investigated via numerical models and finite element analysis. The results are shown in Fig. 3. The microchip contains a section of the straight channel followed by a section of the trident branches (as shown in Fig. 2B), where the acoustic and hydrodynamic mechanisms drive the motion of particles in different ways.

Fig. 3 Design and optimization of the acoustofluidic-based urine microscopic examination chip. A) Numerical simulation of the acoustic pressure distribution and particle tracing results. B) Hydraulic–electric circuit model of the microchannel network. Channels and pumps are analogous to the resistance and constant source in an electric circuit, and fluid flux is analogous to current. C) Illustration of the corresponding microfluidic channel structure. D) Calculation of the fluid rate distribution in each channel according to the hydraulic–electric circuit. E) Numerical simulation of flow rates in the microchip. F) The flow rates in the middle and side channels according to numerical simulation.

The acoustofluidic numerical simulation is conducted using COMSOL Multiphysics. Fig. 3A shows acoustic pressure distribution in the microchip and the tracing of particles in the acoustic field. There is a straight acoustic pressure node line in the middle of the straight channel, which is formed by the resonance of standing acoustic waves reflected by hard channel walls. The resonance frequency is found to be 2.001 MHz. Under the resonance frequency, the acoustic radiation force is maximum, thus particles will be well aligned at the center of the middle channel. It is also noted that there is an obvious wavy distribution of the acoustic field in the middle channel in the section of the trident branches. This phenomenon was also noticed in previous studies.^48–50 The acoustic wave in the middle channel interferes with the wave reflected by the branch and side channels leading to a wavy acoustic field in the middle channel. Although the wavy distribution of the pressure node may be not perfect for aligning, the acoustic radiation force can trap the cells and stop them from entering side channels.

At each trident branch, the hydrodynamic mechanism dominates the motion of fluid. To investigate the distribution of fluid, the hydraulic circuit for the microchannel network is designed using an analogy to an electric circuit.^50,51 Namely, as shown in Fig. 3B and C, the pressure difference ΔP, flow rate Q, and hydraulic resistance R in the microchannel are analogous to the electrical voltage difference, current, and resistance in the conductive wire, respectively. Meanwhile, two pumps at the inlet and outlet are analogous to constant flow sources. The size and hydraulic resistances for all channels are summarized in Tables S1 and S2 (ESI†). The hydraulic–electric analogy calculation is used to analyse the flow state in the channel. The hydraulic–electric model is simple that it does need the process of building geometric models and meshing by simplifying the chip as one electric circuit. Fig. 3D shows that most of the liquid enters the side channels, thus the flow rate in the middle channel decreases while the flow rate in the side channels increases correspondingly. Assuming that all the particles remains in the middle channel, thus the concentration factor of particles (defined as the outlet particle concentration divided by that of the inlet) can be obtained by the flow rate in the inlet divided by that of the middle channel. To minimize the loss of bioparticles, it is crucial to make sure that the concentration factor at each junction is similar.⁵⁰ According to the hydraulic–electric circuit model, the flow rate of the inlet is 1000 μL min⁻¹, and the flow rate in the middle channel is 215.5, 46.2, and 10 μL min⁻¹ at each branch, respectively. As a result, the suspension can be enriched by approximately 4.5 to 4.8-fold at each branch so that the total concentration factor is approximately 100-fold with the least loss of cells.

According to the calculation, the length and width of the branch channels are designed to adjust flow resistance (Fig. S2 and S3, ESI†). The height of the whole channel is 150 μm and the width of the middle channel is 375 μm. The length of the prefocus channel is 28 mm and the length of the middle channel after each junction is 8 mm, 5 mm and 1 mm due to the decrease of the flow rate of the sample in the middle channel.

To verify the calculation by the hydraulic–electric circuit model, numerical fluid simulation is also performed using COMSOL Multiphysics, as shown in Fig. 3E and F. Although a little complicated, the numerical simulation is more versatile for diverse structures especially for some channels whose flow resistance is hard to calculate. The flow rate of the enriched suspension in the middle channel decreases to a quarter to a fifth correspondingly after passing through each junction. The values of the flow rate obtained from numerical simulation matches well with those calculated using the hydraulic–electric circuit model, proving that the geometric design of the microchip is able to fulfil the goals.

Performance of the acoustofluidic microchip

We characterized the concentration effect of the chip, and the results are shown in Fig. 4. Firstly, to illustrate the flow of fluid, we designed a testing microchip with three inlets and injected a fluorescent solution and deionized water into the middle and side inlets with flow rates of 10 μL min⁻¹ and 200 μL min⁻¹. As shown in Fig. 4A, the fluorescent solution is intentionally focused at the middle of the channel by large volume sheath liquid coming from the side inlets. The Reynolds number is low (Re = 55.3) so that these two liquids almost do not mix with each other in the microchip. Therefore, their boundary can reflect the natural liquid streamline in the channel. The share of the fluorescent solution in the middle channel expands after flowing through each junction because a lot of non-fluorescent water enters the side channels symmetrically. Finally, the fluorescent solution occupies the entire middle channel at the outlet region. The results indicate that only a small fraction of liquid will exit from the middle outlet and a high concentration factor can be achieved if the particles are focused in the middle.

Fig. 4 Concentration of particles using the acoustofluidic microchip. A) The flow state of fluorescent solutions (10 μL min⁻¹) and DI water (200 μL min⁻¹) in the channels from the inlet to the outlet. Their boundary can show the natural streamline in the channel. B) Lateral positions of 10 μm particles in each branch of the microfluidic chip. The image frames captured over a certain time period are stacked to show the trajectories of particles. C) The comparison of the concentration factor in each branch among the hydraulic–electric model, numerical simulation and the experimental data. (D) The comparison of the flow rate change in the middle channel among the hydraulic–electric model, numerical simulation and the experimental data.

The concentration performance of the device was tested using polystyrene (PS) beads with a diameter of 10 μm. The chip with a single-inlet channel was used for acoustofluidic focusing of PS beads and urinalysis. The solution containing particles was pumped into the device at a flow rate of 1000 μL min⁻¹. The image frames captured over a certain time period were stacked using the Z projection option of ImageJ software to create stacked composite images proposed in a previous article.³¹ To clearly illustrate the cell distribution, the color of the composite image was inverted using ImageJ software. The results are illustrated in Fig. 4B. Driven by acoustic radiation force, the particles are aligned at the middle of the channel before the first branch. At each junction, the drag force induced by liquid flow will shift the motion of particles off aligned. But owing to the prefocusing effect and the acoustic trapping, the particles still remain in the middle channel without entering the side channels. After flowing through each junction, the flow rate of particles in the middle channel will decrease dramatically and there is enough time for particles to relocate towards the middle of the channel driven by acoustic radiation force. After repeating the process three times, the concentration of particles is boosted dramatically compared to the original sample.

We characterized the concentration factor and flow rate at each branch and compared the results with theoretical data. The number of the particles passing through each branch is obtained by manually counting particles using a CCD camera for 1 minute. The flow rates are calculated by measuring the mean velocity of particles. The experiments were performed when infusing the suspension with a flow rate of 1 mL min⁻¹. As shown in Fig. 4C and D, the concentration factor is 4.5 to 4.8-fold, which matches with the hydraulic–electric analogy model and numerical simulation data. The total concentration factor can be approximately 100-fold. The flow rates in the middle channel also matches the theoretical data.

We also characterized the concentration factor and recovery rate under varied conditions. The results are shown in Fig. 5. Different flow rates and actuation voltage were used to test the performance of the chip and to determine the optimal parameters. First, the actuation voltage and the outlet flow rate of the sample is set to 10 V and 10 μL min⁻¹, respectively. The inlet flow rate is changed from 600 μL min⁻¹ to 1200 μL min⁻¹. As shown in Fig. 5A, when the flow rate is below 1000 μL min⁻¹, the concentration factor increases linearly and the recovery rate remains almost constant. The recovery rate decreases dramatically at a higher flow rate because a lot of particles will flow to the side junctions especially in the third branch. The concentration factor of particles also decreases at a higher flow rate due to the decrease of the recovery rate. Therefore, the optimal flow rate is 1000 μL min⁻¹. The influence of actuation voltage over the transducer on the recovery rate was tested at this flow rate. The results are shown in Fig. 5B. The recovery rates of particles will increase with higher voltage, because the acoustic radiation force increases with the improvement of the voltage. When the voltage is above ∼10 V, the recovery rates reach a relatively constant level. Thus, the voltage is set to 10.2 V, and under these conditions, the recovery rates of 10 and 15 μm particles are 95.9 ± 2.8% and 96.4 ± 1.1%, respectively. Compared with other devices reported in previous research, our chip can achieve both a high throughput and concentration factor (Fig. 5C), which demonstrates superior advantages for urine microscopic examination.

Fig. 5 Characterization of the acoustofluidic-based microchip. A) The total concentration factor and recovery rate of 10 μm particles at different flow rates. The left Y-axis (red) and the right Y-axis (black) represent the recovery rate and concentration factor of particles, respectively. B) The recovery rate of 10 μm and 15 μm particles at different actuation voltages from 6 V to 12 V. C) The comparison of this acoustofluidic-based microchip with devices reported in previous research in terms of the throughput and concentration factor.

Automated sample dispensing and image recognition

Following the urine concentration via using the combination of acoustic and hydrodynamic mechanisms, the enriched sample needs to be dispensed into the observation chamber for microscopic analysis. To complete the whole process automatically, we designed and characterized the sample dispensing module and image recognition algorithm, and the data are shown in Fig. 6.

Fig. 6 The process of automated micro-dispensing and urine sediment microscopic examination based on this system. A) The dispensing of liquid (red ink) in the chamber on the glass slide due to the capillary effect. B) 2 mL native urine can be concentrated using the chip within 2 minutes to obtain a 20 μL enriched sample. C) The images of native and enriched urine samples, showing the presence of RBCs and WBCs. D) The training process of the image recognition classifier. RBCs and WBCs in the training set are distinguished and marked first, followed by detection and segmentation of RBCs and WBCs via the classifier. Finally, the RBCs and WBCs are counted independently. E) The concentration factor and recovery rate of the acoustofluidic-based system for the process of urine.

A microchamber with dimensions of 15 mm by 15 mm was designed at the outlet of the middle channel. The height of the chamber is 80 μm. Fig. 6A shows that the sample can be guided to enter and distribute uniformly in the chamber on the glass slide due to the capillary effect within 120 s. For urine microscopic examination, 2 mL native urine can be processed in 2 minutes to obtain a 20 μL enriched sample for further analysis using our microchip (Fig. 6B). Then the microchip is examined under the microscope. Ten different images for each slide are taken to decrease error and the average value is used as the outcome. The urine sediment mainly contains erythrocytes (red blood cells; RBCs) and leukocytes (white blood cells; WBCs). The concentration of cells in the sample processed by our microchip is increased dramatically (Fig. 6C).

The images taken using the microscope are processed through a machine learning-based image recognition algorithm to quantitatively analyse the cells in the enriched samples. The image recognition module was developed through the Trainable Weka Segmentation plugin of the ImageJ software (https://imagej.net/plugins/tws/). The process of automated image recognition is illustrated in Fig. 6D and S5.† Firstly, RBCs and WBCs in the training set are distinguished and marked by two experienced medical technologists. WBCs usually have larger cell diameters and wrinkled membranes compared with RBCs (as shown in Fig. 6C). Then the classifier can be trained to learn from the marked training data and perform in unknown images later to detect and segment RBCs and WBCs. After being separated by the classifier, RBCs and WBCs are counted independently. The total process includes 425 images as the training set and 110 images as the testing set. The detection accuracy of RBCs and WBCs is 94.6 ± 3.5% and 95.1 ± 1.8%. The accuracy can be improved further by adding more images to the training set in further research.

Finally, we tested the performance of our platform using a human urine sample. As shown in Fig. 6E, the recovery rate of urine cells is 95.6 ± 2.7% and the concentration factor is 95.2 ± 2.1 fold. Therefore, our acoustofluidic-based system can achieve a high throughput and yield simultaneously with a single run. With the ability to process samples with bulk volume and low particle concentration, as well as automated detection, our system is able to perform urine microscopic examination.

Acoustofluidic-based urine examination

Human urine samples were used to test the performance of this system. The performance of our device was validated by comparing with the manual microscopic method, which is currently regarded as the gold standard of urinalysis. Fifty randomly selected patients who attended the laboratory of a hospital were studied. Urine samples were collected from people who attended routine diagnostic tests at The Second Hospital of Dalian Medical University. The collection and use of urine samples were approved by the institutional committee of the Institutional Ethical Committee (IEC) for The Second Hospital of Dalian Medical University (no. 2024-024) and informed consent was obtained from both volunteers and patients. All experiments were performed in compliance with the Chinese laws and following the institutional guidelines. At least 15 mL urine sample was collected for each person and each sample was separated to two parts. Then the samples were processed by conventional manual microscopic examination and our acoustofluidic-based automated examination separately. As illustrated in Fig. 1, the manual examination procedure was taken according to the standard procedure.⁵² 10 mL sample was centrifuged for 5 min at 1500 rpm (400g) for microscopic examination. The supernatant was decanted until 0.2 mL urine remained at the bottom of the tube. Then the sediment was resuspended, and one drop of sediment was placed on a microscope slide, covered with a coverslip, and examined under a microscope. Evaluation of urine formed elements was performed by two experienced medical technologists. During examination, at least 10 different microscopic fields of each slide were scanned and counted at a high power field (400×). The final results were also calculated by averaging the number of cells in each field.

The linear regression line of the WBC and RBC examination results between the automated device and manual microscopy method is presented in Fig. 7A and B. There is good agreement between these two methods based on the Pearson correlation judgement. The coefficient (r) values is 0.968 (95% CI 0.944–0.982) for WBCs and 0.964 (95% CI 0.937–0.980) for RBCs, respectively. The Bland and Altman analysis⁵³ and the comparison between these two methods for RBCs and WBCs are shown in Fig. S7 and S8 (ESI†).

Fig. 7 The comparison of the acoustofluidic-based system and manual microscopy method for urine testing. Linearity regression analysis of the cell counting results for A) WBCs and B) RBCs using this system and the manual microscopy method. C) The images of other biomarkers such as epithelial cells, crystals, casts and bacteria in the concentrated urine sample using this system.

The within-run imprecision of this device was also analysed. Urine samples with low and high concentrations were selected for the test of RBCs and WBCs. And they were processed repeatedly 20 times. The coefficient of variation (CV) for RBCs is 15.2% for a mean concentration of 2.4 cells per μL and it drops to 6.4% for a mean concentration of 44.2 cells per μL. The CV for WBCs is 13.8% for a mean concentration of 2.2 cells per μL and it drops to 5.1% for a mean concentration of 26.7 cells per μL.

Besides counting the number of RBCs and WBCs, our acoustofluidic-based system can also capture other sediments in human urine. Fig. 7C shows the image of epithelial cells, crystals, casts and bacteria found in the human urine. In this paper, the identification process is based on the shape, size and surface topography of the bioparticles by the algorithm. Thus, some biomarkers that have similar size, shape and surface topography to WBCs and RBCs may be misclassified and lead to a false positive result. For example, a few renal tubular epithelial cells may be misclassified as WBCs. A few fungal spores and round crystals may be misclassified as RBCs and lead to a false positive result. These errors can be reduced by enriching the numbers and the categories of the training set. By training the model using images with labelled epithelial cells, crystals and fungal spores, the algorithms can be improved with the ability to distinguish these biomarkers. These components can also be used as valuable biomarkers that are related to various diseases such as inflammation of the urinary tract system, chronic urinary tract infection or kidney damage.^8,9,54,55 Since these urine sediments hardly appear in most urine, the deficiency of microscopy images limits the training process of the module. However with the accumulation of more images, the current machine learning image recognition algorithm can be updated. There is great potential for the automated quantitative analysis of these urine sediments in the future.

Conclusions

In this work, we developed an acoustofluidic-based urine microscopy examination system that can achieve automated microscopic analysis of the urine sediment. Native urine is treated using an acoustofluidic-based microchip with three parallel branch channels that can concentrate the urine sediment such as cells, crystals and bacteria. It can achieve 1000 μL min⁻¹ throughput and 95.2 ± 2.1 fold concentration with a single run. After flowing out of the chip, the enriched sample is dispensed in the chamber on a glass slide based on the capillary force. Then the microscopy images of the enriched sample are processed by a machine learning algorithm, thus RBCs and WBCs can be quantitatively analysed automatically. The detection accuracy of RBCs and WBCs are 94.6 ± 3.5% and 95.1 ± 1.8%. Finally, urine samples from 50 patients were tested and the correlation coefficient between our system and the conventional manual microscopic examination is above 0.96. This work demonstrates a promising method for widespread automated urine analysis and point-of-care urine testing due to its compact structure and small volume. With the help of our system, urinalysis in clinical laboratories can be updated to be more efficient and less labour intensive.

Data availability statement

The data that support the findings of this study are available from the corresponding author upon reasonable request.

Author contributions

Xin He and Feng Ren conducted the experiments and wrote the manuscript; Yangyang Wang and Zhiyuan Zhang performed the chip design; Jiming Zhou, Jian Huang and Shuye Cao provided suggestions on data analysis; Jinying Dong and Renxin Wang distinguished and marked all sorts of cells in the images of training sets; Mengxi Wu performed the review and editing; Junshan Liu supervised the work.

Conflicts of interest

There are no conflicts to declare.

Acknowledgements

This work is supported by the National Key Research and Development Program of China (2022YFB3204600) and the funding from the Dalian University of Technology (DUT23YG215).

References

1. J. R. Delanghe and M. M. Speeckaert, Clin. Biochem., 2016, 49(18), 1346–1350.
2. C. K. Chen, J. Liao, M. S. Li and B. L. Khoo, Theranostics, 2020, 10(17), 7872–7888.
3. C. Cavanaugh and M. A. Perazella, Am. J. Kidney Dis., 2019, 73(2), 258–272.
4. N. Sharda, O. Bakhtar, B. Thajudeen, E. Meister and H. Szerlip, Lab. Med., 2014, 45(4), e152–e155.
5. J. A. Simerville, W. C. Maxted and J. J. Pahira, Am. Fam. Physician, 2005, 71(6), 1153–1162.
6. M. Yuan, Y. Tan, J. Li, X. Yu, H. Zhang and M. Zhao, Int. Immunopharmacol., 2021, 90, 107122.
7. M. A. Perazella, S. G. Coca, I. E. Hall, U. Iyanam, M. Koraishy and C. R. Parikh, Clin. J. Am. Soc. Nephrol., 2010, 5(3), 402–408.
8. M. Kanbay, B. Kasapoglu and M. A. Perazella, Int. Urol. Nephrol., 2010, 42(2), 425–433.
9. L. S. Chawla, A. Dommu, A. Berger, S. Shih and S. S. Patel, Nephron Clin. Pract., 2008, 110(3), C145–C150.
10. C. Hwang, W. J. Lee, S. D. Kim, S. Park and J. H. Kim, Biosensors, 2022, 12(11), 1020.
11. M. Oyaert and J. R. Delanghe, J. Clin. Lab. Anal., 2019, 33(5), e22870.
12. R. Lei, R. Huo and C. Mohan, Expert Rev. Mol. Diagn., 2020, 20(1), 69–84.
13. Z. Zaman, G. B. Fogazzi, G. Garigali, M. D. Croci, G. Bayer and T. Kránicz, Clin. Chim. Acta, 2010, 411(3–4), 147–154.
14. M. Oyaert and J. Delanghe, Ann. Lab. Med., 2019, 39(1), 15–22.
15. T. I. Chien, J. T. Kao, H. L. Liu, P. C. Lin, J. S. Hong, H. P. Hsieh and M. J. Chien, Clin. Chim. Acta, 2007, 384(1–2), 28–34.
16. F. D. Ince, H. Y. Ellidağ, M. Koseoğlu, N. Şimşek, H. Yalçin and M. O. Zengin, Pract. Lab. Med., 2016, 5, 14–20.
17. T. Li, D. Jin, C. Du, X. Cao, H. Chen, J. Yan, N. Chen, Z. Chen, Z. Feng and S. Liu, Comput. Methods Biomech. Biomed. Eng. Imaging Vis., 2020, 8(1), 109–114.
18. C. Kucukgergin, E. Ademoglu, B. Omer and S. Genc, Scand. J. Clin. Lab. Invest., 2019, 79(7), 468–474.
19. G. Previtali, R. Ravasio, M. Seghezzi, S. Buoro and M. G. Alessio, Clin. Chim. Acta, 2017, 472, 123–130.
20. P. Tantisaranon, K. Dumkengkhachornwong, P. Aiadsakun and A. Hnoonual, Pract. Lab. Med., 2021, 24, e00203.
21. R. Palsson, M. R. Colona, M. P. Hoenig, A. L. Lundquist, J. E. Novak, M. A. Perazella and S. S. Waikar, JAMA Netw. Open, 2020, 3(8), e2013959.
22. M. A. Witek, I. M. Freed and S. A. Soper, Anal. Chem., 2020, 92(1), 105–131.
23. R. Nasiri, A. Shamloo, S. Ahadian, L. Amirifar, J. Akbari, M. J. Goudie, K. J. Lee, N. Ashammakhi, M. R. Dokmeci, D. Di Carlo and A. Khademhosseini, Small, 2020, 16(29), 2000171.
24. E. K. Sackmann, A. L. Fulton and D. J. Beebe, Nature, 2014, 507(7491), 181–189.
25. D. Mark, S. Haeberle, G. Roth, F. Von Stetten and R. Zengerle, Chem. Soc. Rev., 2010, 39(3), 1153–1182.
26. A. U. R. Aziz, C. Geng, M. Fu, X. Yu, K. Qin and B. Liu, Bioengineering, 2017, 4, 39.
27. N. Pamme, Lab Chip, 2006, 6(1), 24–38.
28. P. Paiè, T. Zandrini, R. M. Vázquez, R. Osellame and F. Bragheri, Micromachines, 2018, 9, 200.
29. J. Rufo, P. Zhang, R. Zhong, L. P. Lee and T. J. Huang, Nat. Commun., 2022, 13(1), 3459.
30. Y. Fan, X. Wang, J. Ren, F. Lin and J. Wu, Microsyst. Nanoeng., 2022, 8(1), 94.
31. N. Xiang and Z. Ni, Lab Chip, 2022, 22(4), 757–767.
32. C. Liu, J. Guo, F. Tian, N. Yang, F. Yan, Y. Ding, J. Wei, G. Hu, G. Nie and J. Sun, ACS Nano, 2017, 11(7), 6968–6976.
33. S. L. Feng, A. M. Skelley, A. G. Anwer, G. Liu and D. W. Inglis, Biomicrofluidics, 2017, 11(2), 024121.
34. K. Khoshmanesh, S. Nahavandi, S. Baratchi, A. Mitchell and K. Kalantar-zadeh, Biosens. Bioelectron., 2011, 26(5), 1800–1814.
35. M. Wu, K. Chen, S. Yang, Z. Wang, P. H. Huang, J. Mai, Z. Y. Li and T. J. Huang, Lab Chip, 2018, 18(19), 3003–3010.
36. M. Antfolk, C. Magnusson, P. Augustsson, H. Lilja and T. Laurell, Anal. Chem., 2015, 87(18), 9322–9328.
37. T. Zhu, R. Cheng, S. A. Lee, E. Rajaraman, M. A. Eiteman, T. D. Querec, E. R. Unger and L. Mao, Microfluid. Nanofluid., 2012, 13(4), 645–654.
38. M. A. Mahani, A. N. Karimvand and N. Naserifar, J. Sep. Sci., 2023, 46(19), 2300257.
39. X. J. Hu, H. L. Liu, Y. X. Jin, L. Liang, D. M. Zhu, X. Q. Zhu, S. S. Guo, F. L. Zhou and Y. Yang, Lab Chip, 2018, 18(22), 3405–3412.
40. C. Church, J. Zhu, G. Wang, T. R. J. Tzeng and X. Xuan, Biomicrofluidics, 2009, 3(4), 044109.
41. Y. Akiyama, T. Egawa, K. Koyano and H. Moriwaki, Sens. Actuators, B, 2020, 304, 127328.
42. A. Urbansky, F. Olm, S. Scheding, T. Laurell and A. Lenshof, Lab Chip, 2019, 19(8), 1406–1416.
43. W. Zhao, T. Zhu, R. Cheng, Y. Liu, J. He, H. Qiu, L. Wang, T. Nagy, T. D. Querec, E. R. Unger and L. Mao, Adv. Funct. Mater., 2016, 26(22), 3990–3998.
44. N. A. M. Yunus, H. Nili and N. G. Green, Electrophoresis, 2013, 34(7), 969–978.
45. D. Das, K. Biswas and S. Das, Med. Eng. Phys., 2014, 36(6), 726–731.
46. O. Jakobsson, S. S. Oh, M. Antfolk, M. Eisenstein, T. Laurell and H. T. Soh, Anal. Chem., 2015, 87(16), 8497–8502.
47. T. Jonai and Y. Akiyama, Sens. Actuators, B, 2023, 378, 133127.
48. M. Nordin and T. Laurell, Lab Chip, 2012, 12(22), 4610–4616.
49. P. Augustsson, J. Persson, S. Ekström, M. Ohlin and T. Laurell, Lab Chip, 2009, 9(6), 810–818.
50. T. Jonai, Y. Ohori, T. Fujii, A. Nakayama, H. Moriwaki and Y. Akiyama, Sep. Purif. Technol., 2023, 315, 123697.
51. Z. Li, C. Liu and J. Sun, Lab Chip, 2023, 23(15), 3311–3327.
52. T. Kouri, G. Fogazzi, V. Gant, H. Hallander, W. Hofmann and W. G. Guder, Scand. J. Clin. Lab. Invest., 2000, 60, 1–96.
53. J. M. Bland and D. G. Altman, Lancet, 1986, 327(8476), 307–310.
54. M. A. Perazella, Am. J. Kidney Dis., 2015, 66(5), 748–755.
55. M. A. Perazella and S. G. Coca, Clin. J. Am. Soc. Nephrol., 2012, 7(1), 167–174.

---

Footnotes

† Electronic supplementary information (ESI) available. See DOI: https://doi.org/10.1039/d4lc00408f

‡ Xin He and Feng Ren contributed equally to this work.

---