Poster Y7 - Rapid Quantification of CD4+ Immune Cells in Whole Blood Using an Artificial Intelligence-Based Microfluidic Cytometer

Introduction: CD4+cell counts are valuable indicators of disease progression for millions of patients infected with HIV. Flow cytometry is the gold standard for counting CD4+ cell levels; however, due to high cost and the need for trained personnel, flow cytometers are inaccessible to most patients in rural and resource-limited communities. Point-of-care cytometers offer several advantages over traditional flow cytometry regarding cost and ease of use, yet these devices still rely on specialized equipment such as impedance analyzers and fluorescent microscopes, which limit their suitability for resource-constrained settings. Artificial intelligence (AI) techniques have undergone significant evolution in the last few years, enabling the counting of cells in complex heterogeneous populations using computer vision. By tagging CD4+ cells with antibody-coated beads, we enable the easy identification of CD4+ cells that can be detected using AI computer vision to rapidly quantify CD4+ T cell levels from diluted whole blood in a pump-free microfluidic device driven by slug flow.

Materials and

Methods: The microfluidic device was created by CNC milling a polymethyl methacrylate template. Polydimethylsiloxane was then used to generate an inverse template using soft lithography. The mold was bound to a glass slide to produce the microfluidic device. The device consists of a 4 mm long a constriction region that is 2 mm wide and 30 μm thick to enable visualization of passing cells in a single focal plate enabling AI to identify bead bound cells as they pass (Figure 1A). Blood was collected from healthy human volunteers after informed consent using the finger prick method according to an approved Rice University IRB protocol. Bead optimization studies were done by incubating 50 μL of whole blood with varying quantities of CD4 antibody-coated magnetic beads for 5 minutes at room temperature (~23 °C). Samples were washed three times using a magnet to separate bound and unbound cells. Unbound cells were prepared for flow cytometry to validate binding specificity. CD4+ T cells were identified using CD45, CD3, and CD4 antibodies to stain surface markers. The device was loaded with 20 μL of 10X diluted whole blood tagged with beads. As the sample passed through the device, a video was recorded. Frames from the video were extracted, and bound cells were identified using A YOLOv5s model to identify and count CD4+ T cells based on the presence of bound beads. The model was trained and validated on over 4,000 annotated images. These results were compared to CD4+T cell levels quantified using flow cytometry.

Results, Conclusions, and Discussions: Results
The number of CD4 antibody-coated beads needed per sample was determined by adding 1,200,000 to 150,000 beads to whole blood. After incubation, 91-98% of CD4+T cells were bound. The sample with only 150,000 beads resulted in significantly less binding than samples using 1,200,000 and 300,000 beads (Figure 1B). Interestingly, as the number of beads in the sample decreased, more aggregation of cells and beads was noted. In the 300,000 beads sample, clusters of two or more CD4+ cells appear linked by antibody-coated beads. This effect is likely caused by a low bead-to-cell ratio and is resolved by adding more beads, as seen in the samples with 1,200,000 and 600,000 beads, enabling easy detection of singular CD4+cells (Figure 1C). A comparison of samples quantified using the clinical gold standard flow cytometry and the microfluidic device resulted in good agreement between the methods, resulting in an R2 value of 0.8867 (Figure 1D). Further, due to the rapid 5-minute sample preparation and 8-minute quantification, the device reports CD4+ T cell levels in under 15 minutes.

Discussion
This microfluidic device enables the rapid point-of-care quantification of CD4+ T cell counts without the need for a skilled operator or expensive equipment through the integration of AI to rapidly detect bead-tagged cells. The CD4+ T cell counts of healthy volunteers produced by the microfluidic device correlate well to flow cytometry levels. Future studies will explore the accuracy of diseased patient samples and the feasibility of implementation in low-resource settings.

Conclusion
This microfluidic device has the potential to rapidly generate accurate CD4+ T cell counts in under 15 minutes.

Acknowledgements (Optional):