Facile quantification of nanosized bioparticles in bright-field micrograph via deep learning

The drawbacks of laboratory diagnosis that rely on centralized facilities and require significant sample-to-answer time have been highlighted by outbreaks such as COVID-19. Automated rapid/quantitative diagnostic technologies based on intuitive/inexpensive test kits have the potential to diagnose patients immediately at point-of-care and to provide appropriate dosing and treatment. The integration of advanced nanophotonic technologies with artificial intelligence offers promising opportunities for advancing biomedical research and improving disease diagnosis and treatment. Here, we propose a novel approach to quantify nano-sized bioparticles by combining nanophotonics and deep learning (DL). We employed Gires-Tournois (GT) resonators as an immunosensor platform functionalized with antigen-specific antibodies without labelling or amplification. Based on the GT resonator, nanoscale bioparticles are dynamically detected and the number of particles is inferred from optical microscope images by a convolutional neural network as a vision-based DL model. Our DL-based algorithm improves the accuracy of particle evaluation compared to rule-based algorithms by automatically refining surface defects or impurities. Our system demonstrated comparable limits of detection to existing diagnostics and was applied to diverse sizes of analytes using transfer learning, making it a potentially useful tool for screening diagnosis in preparation for emerging viruses.