Stable and High-Throughput Single-Cell Sorting of Food Bacteria Using Spatiotemporal Video-Enhanced Raman Tweezers

Area of Science:

• Microfluidics
• Spectroscopy
• Machine Learning

Background:

• Rapid detection of foodborne pathogens is vital for food safety.
• Raman tweezers spectroscopy (RTS) offers label-free single-cell analysis but struggles with high flow rates causing motion blur and low signal-to-noise ratios (SNR).
• Existing methods are limited in high-throughput inline food inspection.

Purpose of the Study:

• To develop an advanced system for rapid, high-throughput, and accurate detection of microorganisms in liquid food matrices.
• To overcome the limitations of traditional RTS in dynamic, high-flow environments.
• To enhance automated food safety screening using a novel microfluidic and deep learning approach.

Main Methods:

• Integration of a deceleration-optimized microfluidic chip with a deep learning-based visual feedback loop, forming the Spatiotemporal Video-Enhanced Raman Tweezers (SVERT) system.
• Development of a Local-Global Unified Denoising Network (LGU-Net) for high-fidelity bacterial structure recovery from low-SNR video streams.
• Experimental validation using bacterial species relevant to food safety and testing in a commercial beverage sample.

Main Results:

• The SVERT system achieved a deterministic processing latency of ~0.49 ms.
• Optical trapping success rate dramatically improved from 21.27% ± 2% to 91.47% ± 1.8%.
• Achieved 96.74% classification accuracy for four bacterial species and successfully isolated trace E. coli in a commercial beverage.

Conclusions:

• The SVERT system effectively mitigates motion blur and low SNR issues in high-throughput microbial detection.
• This technology significantly enhances spectral acquisition efficiency and bacterial classification accuracy.
• The system demonstrates practical robustness for real-world food safety screening, with potential for broader applications.