Machine-learning-assisted waveguide scattering microscopy for the immunological detection of bovine brucellosis: A proof concept

Abstract

Photonic biosensors based on optical waveguides are at the forefront of biosensing technology, offering exceptional sensitivity and robustness. This study presents a proof-of-concept for detecting Brucella abortus antibodies in bovine serum using He-Ne laser-excited integrated optical waveguides. The antigen-antibody interactions in positive samples resulted in agglutination, forming scattering spots on the light-coupled waveguide, while no such spots were observed in negative samples. These spots were imaged over time using a microscope-coupled camera, generating 718 images that were processed to create a dataset for an artificial neural network (ANN). The ANN accurately distinguished between positive and negative samples, achieving 98.6 % accuracy, 98.7 % precision, and 98.7 % recall, with only a 1.4 % loss. This method detected bacterial antibodies in real animal samples within 20 min, using just 100 µL of reagents without requiring prior waveguide surface modification for antigen immobilization. Combining light scattering-based sensing protocols in photonic waveguides with machine learning tools offers a promising pathway for revolutionizing infectious disease diagnostics.