Machine learning enabled flexible and low-cost plasmonic sensors for point-of-care sensing (Conference Presentation)

Abstract

Methods of detection for key biomarkers in bodily fluids that are specific, low-cost, and non-invasive are in high demand for various biomedical applications. Specifically, field-portable and cost-effective devices which can enable these measurements to be made at home or in the field are crucial for practical and widespread use of these technologies. Plasmonic sensors form an emerging bio-sensor platform that responds to the specific adsorption of bio-molecules via a spectral change in transmission or reflection mode of operation. However, to read and quantify their spectral response, expensive and bulky optics such as broad-band light sources and high resolution spectrometers are typically employed, severely limiting their potential applications in resource-limited settings. In an effort to build low-cost and compact plasmonic readers, we have developed a computational sensing framework that uses machine learning to statistically differentiate the sensor’s spectral response from fabrication related variability and other noise factors, and select the optimal illumination bands for the lowest-possible read-out error. To validate this framework we constructed a low-cost and field-portable plasmonic reader around the optimal illumination bands selected for different plasmonic nano-hole array designs. We then validated the superior performance of our computational reader by measuring a large number of independently fabricated flexible plasmonic sensors made using scalable, nano-imprint lithography methods without the use of a clean room. Additionally, these structures can subsequently be transfer-printed onto disposable, wearable platforms where they can be chemically modified to specifically and sensitively capture target biomarkers in bio-fluids e.g., tear or sweat, enabling new applications in point-of-care diagnostics.