AI100 digital scanner
The in-house built digital scanner was used to acquire / record images of the GSC filled with the raw urine samples. The digital scanner consisted of the following hardware components:
-
Computer system: A mini-ATX motherboard with Intel i5 quad-core processor, 8 GB RAM, NVIDIA GPU with 4G GPU memory, running Ubuntu Linux (v20.04).
-
Optics system: An optical tube (40× Plan Achromat objective and 10× eyepiece) and an Abbe condenser with a white LED source. A 13 MP USB 3.0 colour camera procured from e-con Systems, Inc., USA, was used. The camera model See3CAM_CU135 contained a 1/3.2 AR1335 CMOS image sensor from ON Semiconductor.
-
XYZ slide stage: The XYZ platform was built using commercially available low-cost ball screws, stepper motors, and some machined parts.
Gravity sedimentation cartridge (GSC)
The GSC comprises a polymethyl methacrylate (PMMA) slide, spacer tape and cover slip. PMMA slides were manufactured by the injection molding method. These slides are optically transparent and have a mirror finish surface with a dimension of 76 × 26 × 1 mm3 (L x W x T). Spacer tapes of thickness ranging from 50 µm to 570 µm were procured from Nitto Denko Corporation (Japan), Teraoka Tapes (Japan), and Avery Dennison (USA) to make different chamber depths. Two coverslips (Microcil Manufacturers, India) of dimensions 22 × 22 mm2 and 22 × 50 mm2 were used to cover the chambers.
Pre-stained slide dye composition
The dye solution consists of 10% crystal violet in methanol (Qualigen, India) and 0.1% tween 20 (SD fine, India). Crystal violet (Nice Chemicals, India) is a dye that stains the cell wall and its nucleus. When dissolved in water, the dye dissociates in the solution to acquire a positive charge that binds to the negative charge in the nucleus and cell wall. The purpose of adding methanol (SD fine, India) to the solution during dip coating was to expedite the evaporation process. Tween 20 was required for the evenly spreading of solution over the slide.
Sample distribution
The midstream urine sample was collected in a sterile container. The sample was processed within 4 hr after the collection and transported at room temperature. A total of 413 fresh abnormal urine samples (Human urine samples) were collected for the analysis. The lab-on-chip cartridge analysis employed 413 samples, whereas the AI model and tele reporting analysis utilized 240 samples. The Institutional Review Committee of Father Muller Medical College Hospital, Mangalore, India, approved all the experimental protocols. The authors ensure that all procedure outlined in this manuscript were conducted following appropriate guidelines and regulations. Prior to sample collection, necessary consent was obtained from the patients.
Figure 1 demonstrates the urine sample distribution. Figure 1(a) presents the distribution of RBC and WBC in different grades, which are NS (Non-significant) (0–5), 1+ (6–10), 2+ (11–20), 3+ (21–50) and 4+ (> 50). The number of RBC samples belonging to different grades are 326, 18, 12, 32 and 25, whereas, for WBC, these are 283, 36, 29, 42 and 23. Figure 1(b) shows the sample distribution of bacteria in urine samples collected. The grade-wise grading for bacteria is NS (0–1), 1+ (1–2), 2+ (3–5), 3+ (6–10), and 4+ (> 10). The number of bacteria samples in this category are 27, 63, 70, 41 and 212. The remaining urine elements epithelial cells, crystals, cast and yeasts are classified as detected or not detected. Figure 1(c) depicts the sample distribution of yeast, epithelial cells, crystals and cast in the urine sample collected. In the not detected category, their numbers are 15, 319, 350 and 359, whereas in the detected category, numbers are 398, 94, 63 and 54.
Cartridge Design & Fabrication
Figure 2 represents the GSC. The cartridge consists of a PMMA slide, spacer tape and coverslip. The PMMA slide and coverslip are cleaned using soap solution (5% Extran (Merck, Germany)). Both are then ultrasonicated in a soap solution (5% Extran), DI water for 5 min and air-dried. Further, the PMMA slide is spray-coated with P100 (Joninn, Denmark). Finally, the coverslip and coated PMMA slide are used directly for the assembly. The PMMA slide forms the base of the cartridge. The spacer tape (25.5 × 17.8 × 0.4 mm3) is placed over the PMMA forming the lateral walls of the chamber. The spacer tape of the required dimension is cut using a laser machine (SM Laser Technology EC6.4, India). The chamber formed by the spacer tape is enclosed from the top using a coverslip in such a way that the inlet / outlet ports are open to the atmosphere. The entire fabrication is carried out in a dust-free area. After loading the raw urine sample at the inlet, it fills the chamber by capillary force. The urine cells and particles are counted / scanned within the scan zone present in the centre of the chamber.
Effect of depth on particle and cell enrichment
The chamber depth is the characteristic dimension that influences the enrichment of cells and particles, and the depth is proportional to the chamber volume. The number of particles and cells per FOV either increases or decreases accordingly. Hence, optimising the necessary depth of the chamber is important to obtain accurate cells and particles per FOV that agree with the gold standard. A suitable grade of urine sample has to be chosen to understand the cells / FOV dependence over the depth of the chamber. Initially, a medium-sized grade sample was used to evaluate the chamber depth significance. The sample chosen for the study was RBC (2+). Experiments were carried out with different depths of the capillary chamber. Figure 3 shows the dependence of cells and particles per FOV over the depth of the capillary chamber. The plot shows a linear dependence. Hence, increasing the depth of the chamber increases the number of cells and particles per FOV and vice versa. The 2 + grade corresponds to the 11–20 cells / FOV. Figure 3 shows that for 11–20 cells / FOV, the depth of the chamber is in the range of 200–400 µm. As the grade-wise analysis of urine cells and particles not corresponding to the absolute number of cells / FOV, the depth of the chamber has a wide range.
If a high-grade urine sample is chosen, i.e. 4+ (> 50 cells / FOV, Supplementary Table S2). Since, the upper end of the grade, i.e. cells / FOV, is not fixed, the one end of the grade in the graph corresponds to infinite chamber depth. Similarly, for low-grade urine samples, i.e. NS (0–5 cells / FOV), the lower end of the grade, i.e. 0 cell / FOV, the one end of the grade in the graph corresponds to the zero chamber depth. Hence, it is not possible to arrive at a definite chamber depth. Thus, low and high-grade samples are unsuitable for accurately determining the chamber’s depth. Hence, the medium-grade urine sample was used for the study. The chamber depth of 400 µm was considered for further studies.
Sedimentation time
When the raw urine sample flows into the chamber by capillary force, the particle distribution is divided into three regions. Figure 2(a) shows the particle distribution in the chamber. The inlet region is called the coarse region, the central region is the focus region, and the outlet region is called the sparse region. The enriched layer, defined by the plurality of the cells and particles in the chamber, includes a coarse region, defined as the group of large-size cells and particles. Further, the enriched layer, defined by the plurality of cells and particles, includes a focus region, defined by the combination of a group of small, moderate and large-size cells and particles. This region is suitable for urine cells and particles counting / scanning as it contains a mixture of all size cells and particles. The area considered for counting / scanning is called a scan zone, which is of size 6 x 8 mm2. Cell counting is avoided close to the chamber's walls as more cells and particles accumulate due to the no-slip condition. Furthermore, the enriched layer includes a sparse region defined by a group of lower-size cells and particles in the plurality of cells and particles. Hence, cells and particle enrichment is based on gravity sedimentation, whose velocity is given by particle settling velocity.
The particle settling that occurs under the influence of gravity can be predicted by Stoke's law21,22. Stock’s law predicts the settling velocity of small spherical parts in a fluid where particles are well separated, i.e. dilute solution, and the fluid behaves like a continuous medium. When a particle falls through a fluid, the interplay between gravity and drag force decides the settling behaviour of the particle. The gravity acting on the particle accelerates, and simultaneously, the drag force deaccelerates the particle. The terminal velocity is given as
V = (2 (ρp -ρf)g r2 ) / (9µ) (1)
Where, V represents the settling velocity of the particle in micrometre per second (µm / s). ρp denotes the density of the particle in grams per millilitre (g / ml), while ρf represents the density of the fluid in grams per millilitre (g / ml). The variable r represents the particle's radius in micrometre (µm), and µ signifies the viscosity of the fluid in centipoise (cP).
Table 1
Settling velocity of cells present in the raw urine samples after loading in GSC.
Urine elements
|
Density (g / ml)
|
Average density (g / ml)
|
Size (µm)
|
Average size (µm)
|
Viscosity (cP)
|
Settling Velocity in a raw urine sample (V µm / s)
|
RBC
|
1.11
|
|
6–8
|
7
|
-
|
2.47
|
WBC
|
1.08
|
|
12–20
|
16
|
-
|
8.71
|
Bacteria
|
1.11
|
|
1–2
|
1.5
|
-
|
0.11
|
Yeast
|
1.11
|
|
3–4
|
3.5
|
-
|
0.63
|
Raw urine sample
|
1.005–1.030
|
1.02
|
-
|
-
|
1
|
-
|
Table 1 represents the settling velocity of cells and particles present in raw urine samples after filling inside the chamber by capillary force. The settling velocity indicates that all the cells and particles settle at the bottom surface of the chamber. The density of all the cells and particles is higher than the raw urine, which gives a positive settling velocity thereby, cells and particles settle at the bottom surface of the chamber and form an enriched layer. The degree of settling depends on the viscosity of the raw urine sample, gravity acting on cells and particles and its size.
Here, the sedimentation time for urine cells and particles was determined for the GSC by analysing it under the microscope at 40x objective. To analyse the sedimentation time, the GSC was visualized under the microscope for a period of 10 min immediately after loading the raw urine sample. A total of 5 samples were studied. Figure 4 shows the images of urine cells and particles in GSC from the time period of 1–10 min. The majority of cells are not in focus till 4 min. However, after 5 min, most of the cells are in focus as they settle down at the bottom surface and form an enriched layer. The GSC cells and particles count data after 5 min agrees with the wet mount shown in supplementary Table S1. Thus, the sedimentation time required for enrichment in GSC is 5 min.
In GSC, the large cells and particles like RBC, WBC, epithelial cells, crystals etc. have relatively higher settling velocity which settles down at the bottom of the chamber within 5 min. While, the small size cells, like bacteria and yeast, have a smaller magnitude of settling velocity and hence, take ~ 60 min and ~ 10 min to settle down at the bottom surface. This occurs only when all the small cells and particles are at the top of the chamber. However, when the well-mixed urine sample is filled in the chamber by capillary force, the cells and particles are distributed throughout the depth of the chamber. Hence, the layer of bacteria and yeast present close to the bottom surface approaches the settled larger urine elements layer within 5 min and forms an enriched layer. To capture urine elements, the AI100 scanner was programmed to acquire images at multiple depths, which are later stacked together to form a single image. The scanner initially focuses and captures the larger particles close to the bottom surface. Then the images of bacteria and yeast above and below the larger particles at a distance of 2 µm are acquired. These three images are stacked together to form a single image for the tele reporting and AI model analysis. Similarly, during the manual microscopy of GSC, fine focusing was used to capture bacteria and yeast after the larger particles are focused.
Automated digital analyzer (AI100)
The AI100 is an imaging-based autoanalyzer that accepts the GSC and scans the sample. The scan was done by capturing twenty-five microscopic images at different locations along the length and width of the scan zone in the cartridge. Thereafter, these images are pushed to the cloud, where a deep learning-based object detection model is invoked. Object detection models are deep models based on convolutional neural networks. The deep layers help in learning from simple to complex features present in the data. Usually, objection detection is done via two types of models, single-stage detection or two-stage detection. The model is trained using a single-stage detector as it suits the inference time feasibility and computes resource bandwidth. The YoloX model23 was chosen as it is one of the state of art single-stage object detectors in computer vision. Before training, labelled data was prepared by getting it annotated via a pathologist. Then the prediction was performed after the model had been trained using annotated data. The model generates potential bounding boxes along with respective confidence scores. This helps in locating medically significant cells in the image. Figure 5 shows the cells located by the AI model over the microscopic image acquired by the AI100. The AI model locates and identifies the various urine cells and particles.
A consolidated report is published once the model is run on all the images captured for a sample. The model currently detects white blood cells (WBC), red blood cells (RBC), epithelial cells, yeast, crystal, cast and bacteria. It gives a quantitative assessment for RBC, WBC and bacteria and a qualitative assessment for epithelial, yeast, cast and crystal. The reporting methodology for AI100 for these cell types is according to the supplementary Table S2. Hence, for RBC, WBC and bacteria, it computes the average cell counts over all the FOV images and grades it as per supplementary Table S2. For qualitative assessment, epithelial, cast, crystal and yeast are marked as detected if detected by the AI model otherwise marked as not detected. Apart from the report generation at the web-based application, the pathologist can even view the microscopic image of urine elements to do tele reporting studies.
Workflow
Figure 6 shows the workflow of GSC. Initially, 100 µl of raw urine sample is loaded into the chamber using a pipette allowing the cells and particles to settle down within 5 min. Here, the cells and particles settle down due to the low viscosity and density of the sample, enriching them at the bottom chamber surface. Finally, the cartridge is loaded into the AI100 device for automated image processing and analysis.