Platform developments of integrated optofluidic imaging for new-generation cytometry

Abstract

The growing interest in personalized medicine is driving the innovative development of biological assays that can detect biomarkers indicative of health and disease at high throughput without sacrificing the measurable information content, which often determines assay specificity. However, available bioassays are challenged by a common trade-off between data content and throughput. The notable example is imaging flow cytometers, which aims to gain spatial respresentation of the cells at the expense of measurement throughput in comparison with standard flow cytometry. Another common challenge in the vast majority of bioassays comes with the use of cytological staining/fluorescence labeling. While these methods provide exquisite assay specificity, they have many limitations, including laborious and costly labelling protocols that perturb the biological samples. Motivated by these challenges, this thesis work aims to develop a high-throughput imaging flow cytometer that enables in-depth label-free assays at the single-cell precision. Here we developed high-throughput optical imaging enabled by an ultrafast all-optical laser-scanning concept, called optical time-stretch. It combines with microfluidic technology to establish an ultrafast optofluidic platform for imaging fast-flowing cells and particles in a microchannel. Furthermore, the image contrast of the label-free cells are enhanced by incorporating quantitative phase imaging (QPI) concept with time-stretch imaging. In this thesis, we adopt an interferometry-free approach for enabling time-stretch QPI, called multi-ATOM. This technique is employed for its unique ability to estimating the optical phase without compromising imaging throughput and resolution. By analyzing quantitative phase information together with spatial content in the image, intrinsic cytological quantities representing cell phenotypes and status are revealed without the use of labeling. Time-stretch imaging provides an ultrafast image generation rate, albeit lacking an image processing mechanism matching the speed. Machine learning is a growing field that is advantageous in accurately classifying input data permitted by extensive training on learning algorithm beforehand. By implementing machine learning into imaging hardware, we developed an integrated computing system for high-speed image processing streamlined with ultrafast multi-ATOM. In our implementation based on a reconfigurable field-programmable gate array (FPGA) system, we demonstrated that the image processing and classification latency outperform the state-of-the-art optofluidic imaging systems. Image-based classification information obtained from the machine learning algorithm can be harnessed for sorting/enriching targeted cells on-demand for downstream bioassays in real-time. We developed a microfluidic cell-sorting device that is based on piezoelectric traduction in the microfluidic channel. This device allows precision actuation of target cells away from the original streamline to the designated outlet at a throughput of 100 cells/sec. Built upon custom design and fabrication steps, we perform a series of proof-of-concept demonstration to show that this microfluidic cell-sorter, activated by fluorescence signal, achieve the high precision and cell viability required by typical bioassay applications. In summary, we propose a new generation of optofluidic imaging flow cytometry that combines label-free ultrafast optical time-stretch imaging, real-time hardware-implemented image-based machine learning classification, and microfluidic cell sorting. This unprecedented combination will usher in new approaches for cost-effective cell-based assays that could impact disease diagnosis and facilitate prognosis management.