In-vitro blood purification using tiny pinch holographic optical tweezers based on deep learning

In-vitro blood purification is essential to a wide range of medical treatments, requiring fine-grained analysis and precise separation of blood components. Despite existing methods that can extract specific components from blood by size or by magnetism, there is not yet a general approach to efficiently filter blood components on demand. In this work, we introduce the first programmable non-contact blood purification system for accurate blood component detection and extraction. To accurately identify different cells and artificial particles in the blood, we collected and annotated a new blood component object detection dataset and trained a collection of deep-learning-based object detectors upon it. To precisely capture and extract desired blood components, we fabricated a microfluidic chip and set up a customized holographic optical tweezer to trap and move cells/particles in the blood. Empirically, we demonstrate that our proposed system can perform real-time blood fractionation with high precision reaching up to 96.89%, as well as high efficiency. Its scalability and flexibility open new research directions in blood treatment.