Imaging Flow Cytometry at >13K events/s Using GPU-Accelerated Computer Vision

• Main Authors: Vedhanayagam, Arpith, Basu, Amar S.
• Format: Conference Proceeding
• Language: English
• Subjects: computer vision; GPU; imaging flow cytometry
• ISSN: 2168-9229
• DOI: 10.1109/SENSORS43011.2019.8956759

Flow cytometers are widely used to rapidly measure characteristics of single cells. Typical laser-based instruments provide throughputs of >10,000 events/s; however, the number of measured features is typically small and apply to the entire cell volume. Imaging flow cytometers (IFC) rely instead on 2D images of the objects, providing hundreds to millions of spatially resolved features. However, the throughput of IFCs is typically lower (several thousand events/s) due to the computational overhead of 2D image processing. Here, we demonstrate a GPU-accelerated computer vision analyzer which substantially increases computational throughput. When coupled to a 300 frame per second (fps) real-time camera, the system is limited by the camera and analyzes 1260 particles/s in a 500x700 pixel video with 4-5 particles/frame. When reading from a solid state disk, the throughput increases to 4500 fps with ~3 particles per frame, resulting in a throughput of 13,500 particles/s. The reported throughput is 2.5-4X higher than existing technologies, paving the way for ultra-high throughput IFC.