Combining Silicon Photonics and Machine Learning for Red Blood Cell Characterization

Abstract

Tens of millions of units of blood are transfused worldwide each year, with each individual unit requiring manual typing by a trained technician prior to transfusion. Several tools exist to help expedite the typing process, yet typing still remains a slow and costly process. In an effort to increase throughput and decrease costs of blood typing, recent work involving silicon photonic biosensors has demonstrated their potential as a rapid, low cost tool for typing blood. This thesis focuses on two separate aspects of silicon photonic blood typing: photonic sensor selection and validation along with automation of downstream data processing. Transverse electric and transverse magnetic mode microring resonators are compared for serologic and phenotypic typing assays. Phenotypic typing data from multiplexed photonic blood typing chips is used with several machine learning algorithms to predict blood types with accuracies rivaling by-hand analysis of the same data.