Deep learning integrated plasmonic electrochemical sensing for fast and accurate pathogen detection

The development of plasmonic electrochemical biosensors using the new generation of deep learning algorithms is a potent pathway toward the troublesome, immediate and field-mediable diagnostics of the pathogen. This article provides a combination of a MobileNet-Transformer and Gated Recurrent Unit (GRU) deep neural network with a nanostructured plasmonic biosensor designed to sense Escherichia coli, Salmonella typhimurium, and Staphylococcus aureus at an early stage. To augment the charge-transfer kinetics in the biosensor, localized surface plasmon resonance (LSPR) is utilized by use of gold-nanoparticle graphene oxide hybrid nanocomposites which lead to maximized electrochemical responses. The platform has ultra-low E. coli, Salmonella, and S. aureus limits of detection of 0.12 pg/mL, 0.17 pg/mL and 0.21 pg/mL, respectively using 5 μL of sample and a time of assay of less than 10 min. The deep learning pipeline processes raw voltammetric signals automatically with MobileNet-Transformer being helpful to determine the features effectively and GRU to reduce the noise related to time. The system was better than baseline CNN and RNN models, with a classification accuracy 95.6% and area under the curve of 0.986 as well as better precision-recall profiles. The vehicular combinations of plasmonic enhancement and deep learning deposition make it possible to realize real-time on-device decision-making that can be made applicable in food safety checks, environmental or point-of-care diagnoses. This paper illustrates a scalable path to AI-assisted electrochemical biosensing and a similar performance on par with laboratory benchtop systems and that is fully compatible with low-cost diagnostic hardware in a portable format.