PYR Health

Overview

PYR Health was a bioengineering and health AI project focused on improving outpatient chemotherapy monitoring. Chemotherapy can cause severe blood-related toxicities, including anemia, neutropenia, platelet deficiencies, and immune suppression. These side effects are dangerous when patients are not monitored frequently, especially in regions where CBC testing is expensive, slow, or inaccessible. The core idea was to build a portable system for low-cost, at-home blood count monitoring that could integrate into physician workflows.

The proposed device combined a disposable microfluidic chip with a low-cost microscopy system and computer vision pipeline. A patient would provide a small finger-prick blood sample, the chip would prepare and separate the sample, and the microscope would image blood cells for automated analysis. The system was designed to generate RBC, WBC, platelet, and WBC differential counts while also capturing cell morphology for deeper clinical insight.

On the technical side, we designed the device around three connected systems: microfluidic sample preparation, low-power brightfield microscopy, and ML-based cell detection. The chip used acoustic microstreaming to separate red blood cells, white blood cells, and platelets before imaging. The vision pipeline used YOLO-style object detection and ensemble computer vision models to identify cell types, estimate densities, and support downstream CBC analysis.

The broader vision was to turn routine CBC monitoring into a remote patient monitoring platform. Instead of relying on infrequent clinic visits, patients could generate longitudinal blood composition data from home. Physicians could use that data to adjust dosing, detect adverse effects earlier, and understand how individual patients respond to chemotherapy over time. Over the long run, the same infrastructure could support population-level cancer care datasets in regions where clinical registries are limited.

What made the project meaningful to me was that it sat at the intersection of hardware, biology, machine learning, and care delivery. The hard problem was not simply building a classifier. It was designing an end-to-end system that could work in real clinical settings: cheap per-test cost, small sample volume, low error rates, easy patient usage, and integration with the realities of outpatient oncology care.

Technical Highlights

• Microfluidic chip design for automated blood sample preparation
• Low-cost brightfield microscopy system
• YOLO-based blood cell detection and classification
• RBC, WBC, platelet, and WBC differential estimation
• Longitudinal chemotherapy toxicity monitoring
• Remote patient monitoring platform for oncology care
• Finalist at UC Big Ideas