Transfer Learning Empowered Multiple-Indicator Optimization Design for Terahertz Quasi-Bound State in the Continuum...

Area of Science:

• Biomedical Engineering
• Nanotechnology
• Optics

Background:

• Terahertz metasurface biosensors utilizing quasi-bound states in the continuum (QBIC) enable label-free, rapid, and ultrasensitive biomedical detection.
• Deep learning has advanced metasurface design, but existing methods often overlook trade-offs between key performance indicators.
• Optimizing multiple indicators like Quality (Q) factor, Figure of Merit (FoM), and Effective Sensing Area (ESA) simultaneously is crucial for superior biosensor performance.

Purpose of the Study:

• To develop a novel transfer learning approach for optimizing multiple performance indicators in terahertz metasurface biosensor design.
• To achieve comprehensive optimization of Q factor, FoM, and ESA for enhanced biosensing capabilities.
• To demonstrate a new paradigm for designing highly sensitive and efficient metasurface biosensors.

Main Methods:

• A two-stage transfer learning strategy was employed, involving pre-training on low-dimensional datasets and fine-tuning on high-dimensional tasks.
• Frequency shift was utilized as a unified criterion to quantify the contribution of each performance indicator.
• The method was validated by designing a biosensor for homocysteine detection.

Main Results:

• The transfer learning method achieved simultaneous optimization of Q factor (26.09%), FoM (48.42%), and ESA (25.49%).
• Data requirements were reduced by 50% compared to conventional deep learning methods.
• The designed biosensor demonstrated homocysteine detection at the ng µL⁻¹ level, with experimental results aligning with theoretical predictions.

Conclusions:

• This study presents a pioneering method for multi-indicator optimization in metasurface biosensor design using transfer learning.
• The developed approach significantly enhances biosensor performance and reduces data demands.
• This work paves the way for advanced trace biological detection technologies.