High Spatiotemporal Precision Mapping of Optical Nanosensor Array Using Machine Learning.

Area of Science:

• Nanotechnology
• Biosensing
• Machine Learning

Background:

• Optical nanosensors, like single-walled carbon nanotubes (SWCNTs), offer high sensitivity for real-time, single-molecule detection.
• Environmental factors (fluid flow, mechanical stress) and low analyte concentrations often limit nanosensor accuracy and detection capabilities.
• Existing methods struggle to achieve optimal limits of detection (LOD) and can misinterpret data due to noise.

Purpose of the Study:

• To develop a machine learning-based single-pixel mapping technique for optical nanosensor arrays.
• To enhance the spatiotemporal precision and sensitivity of nanosensor measurements, particularly for analytes below the LOD.
• To differentiate true analyte signals from environmental noise for more reliable data.

Main Methods:

• Utilized a near-infrared fluorescent SWCNT nanosensor array to measure spatial sensing images at various analyte concentrations.
• Applied machine learning to extract single-pixel features (entropy, Laplacian, neighboring values) from noisy sensor data.
• Trained an AI model to identify specific analyte reaction pixels and distinguish them from noise sources like fluid dynamics or mechanical modulations.

Main Results:

• Achieved a 13-fold increase in detection sensitivity compared to the original LOD.
• Reduced detection time for reporter pixels by half, with F1 scores exceeding 0.9.
• Successfully isolated specific analyte responses from background noise, enabling accurate spatiotemporal reporting.

Conclusions:

• The developed AI-powered single-pixel mapping technique significantly enhances the sensitivity and accuracy of optical nanosensors.
• This method overcomes limitations posed by environmental noise and low analyte concentrations, improving LOD.
• The approach is broadly applicable to various optical nanosensor materials and analytes, advancing applications in diagnostics and analysis.