SLE diagnosis research based on SERS combined with a multi-modal fusion method

Area of Science:

• Biomedical Engineering
• Spectroscopy
• Artificial Intelligence

Background:

• Raman spectroscopy shows promise for disease diagnosis but faces challenges with weak signals in systemic lupus erythematosus (SLE).
• Surface-enhanced Raman spectroscopy (SERS) improves signal but is sensitive to impurities and signal degradation.

Purpose of the Study:

• To develop a fusion technique integrating origin Raman spectral data (ORS) and SERS data for enhanced SLE diagnosis.
• To improve classification accuracy by leveraging the complementary nature of ORS and SERS signals.

Main Methods:

• A two-branch residual-attention network (DBRAN) was proposed for multimodal spectral data fusion.
• Feature extraction utilized residual modules, and attention modules were incorporated for efficient weight allocation of modal features.

Main Results:

• Both low-level and intermediate-level fusion methods significantly improved SLE classification accuracy compared to single modalities.
• The DBRAN intermediate-level fusion achieved 100% accuracy, sensitivity, and specificity for SLE classification.
• Accuracy improved by 10% and 7% compared to ORS unimodal and SERS unimodal modalities, respectively.

Conclusions:

• Multimodal spectral data fusion using DBRAN enables rapid and accurate diagnosis of SLE.
• This approach provides a valuable reference for Raman spectroscopy applications and future clinical translation.