Machine learning unveils surface refractive index dynamics in comb-like plasmonic optical fiber biosensors

Aug 5, 2026

Precise surface refractive index (RI) monitoring underpins biosensing by revealing how bioreceptors interact with analytes. Correlating complex spectral features with RI variations remains difficult in gold-coated tilted fiber Bragg grating (Au-TFBG) surface plasmon resonance sensing, especially when spectral comb modes evolve and transition smoothly. A machine learning regression framework is introduced by combining a regression model with gold-coated tilted fiber Bragg grating sensors to enhance stability and precision, enabling direct correlation between spectral shifts and RI changes across multiple acquisition systems and supporting real-time RI monitoring in insulin functionalization and detection.

LanguageEnglish

CategoryResearch & Report

FormatPDF

Pages9

Reading Time23 min

AccessFree to download

Background and Motivation

Importance of surface RI in biosensing

Optical fiber SPR platforms and Au-TFBG properties

Spectral-comb demodulation and RI-to-spectrum challenges

Proposed Machine Learning Approach

Regression model with gold-coated tilted fiber Bragg grating sensors

Correlating spectral shifts with RI variations

Avoiding individual sensor calibration

Validation and Applications

Robustness across different data acquisition systems

Insulin functionalization and detection experiment

Advancement toward real-time RI monitoring

[https://doi.org/10.1038/s44172-024-00181-9](https://doi.org/10.1038/s44172-024-00181-9) Machine learning unveils surface refractive index dynamics in comb-like plasmonic optical ﬁ ber biosensors Check for updates | Hadrien Fasseaux 1 , Médéric Loyez 1,2,3 & Christophe Caucheteur1,3 | | | --- | --- | | The precise measurement of surface refractive index changes is crucial in biosensing, providing insights into bioreceptors–analytes interactions. However, correlating intricate spectral features, with these refractive index variations remains a persistent challenge, particularly in optical ﬁber gratingsbased Surface Plasmon Resonance sensing. Here, we introduce a machine learning-based approach to address this ongoing issue. We integrate a regression model with gold-coated tilted ﬁber Bragg grating sensors. This enhances signal stability and precision, enabling a correlation between spectral shiftsand refractive index changes. Our approach eliminates the need for individual sensor calibration, thereby bolstering the effectiveness and efﬁciency of the sensing layer. We demonstrate the model’s versatility by showcasing its efﬁcacy across two data acquisition systems with different resolutions, allowing for comparative analysis and robustness enhancement. Its application in a biosensing experiment for insulin functionalization and detection, demonstrates how this breakthrough approach marks an advancement in real-time refractive index monitoring. | | | Surface refractive index (RI) variation plays a crucial role in the realm of biosensing, offering valuable insights into the complex interplay between bioreceptors and their corresponding analytes1–4. Surface plasmon resonance (SPR) has emerged as a highly-responsive technique for monitoring biomolecular interactions, owing totheRI variations occurring in proximity to a metal surface upon biomolecule adsorption5–9.
In recent years, a diverse range of physical platforms grounded in SPR principles, such as the Kretschmann prism or optical ﬁbers, have been extensively explored, leading to the development of commercially viable devices10–16. Optical ﬁber-based platforms offer a compelling array of attributes, including immunity to electromagnetic interference, lightweight and compact design, cost-efﬁciency, wavelength multiplexing capacity, and adaptability17–20.
Among this array of conﬁgurations, gold-coated tilted ﬁber Bragg gratings (Au-TFBG) emerge prominently, distinguished by their temperature compensation ability21. Furthermore, their straightforward fabrication not only preserves the inherent optical properties of the single-mode ﬁber platform but also enables the extraction of cladding modes involved in core mode coupling, achieved through a single spectral measurement22–26. These cladding modes are illustrated in Fig. 1a through the insertion loss spectrum (ILs) associated with | transverse electric (TE) polarization. When operating in the transverse magnetic (TM)-polarization mode27, this dense spectral comb reveals a distinctive notch centered around the resonance wavelength, signifying the region of highest sensitivity (see Fig. 1b) .
In response to these spectral changes prompted by a shift in RI, several demodulation techniques have been conceived, linking shiftsin wavelength or amplitude to corresponding RI alterations23,28–32. Understanding the evolution of the spectrum, characterized by distinct peaks in the spectral comb, is paramount. Each mode undergoes a sigmoidal evolution, with the region of highest sensitivity conﬁned to a speciﬁc range of refractive indices28. Challenges arise when observing RI changes over extended intervals and transitioning smoothly between modes during analysis. However, certain methodologies, such as the cross-demodulation method (refer to “Spectral shape characteristics”), mitigate the impact of such mode transitions.
Furthermore, recent investigations by Chubchev et al.33 have laid the groundwork for integrating mach

FAQs

Why is linking spectral features to surface refractive index changes challenging in Au-TFBG SPR biosensors?

Au-TFBG sensing produces a dense spectral comb whose modes evolve sigmoidally and can transition smoothly as RI changes, making it hard to maintain a consistent mapping between spectrum and RI over extended intervals.

What does the proposed machine learning method change compared with traditional demodulation?

It uses a regression model integrated with gold-coated tilted fiber Bragg grating sensing to correlate spectral shifts directly with RI changes, improving stability and precision and removing the need for per-sensor calibration.

How is the approach validated and where is it demonstrated?

The model’s versatility is shown across two data acquisition systems with different resolutions, and it is applied in a biosensing experiment involving insulin functionalization and detection to support real-time refractive index monitoring.

Aug 5, 2026

Upload by River Wang