A rapid diagnostic approach for COPD utilizing multimodal serum spectra integrated with machine learning algorithms.

Area of Science:

• Biochemistry
• Medical Diagnostics
• Spectroscopy

Background:

• Chronic Obstructive Pulmonary Disease (COPD) is a leading cause of global mortality and disability.
• Current diagnostic methods for COPD lack standardization and molecular insight.
• There is a critical need for efficient and convenient diagnostic tools for COPD.

Purpose of the Study:

• To evaluate serum fluorescence (FS), Raman (RS), and surface-enhanced Raman spectroscopy (SERS) combined with machine learning for diagnosing COPD.
• To compare the diagnostic potential of FS, RS, and SERS in differentiating COPD from non-COPD and healthy individuals.
• To identify optimal spectroscopic and machine learning combinations for accurate COPD classification.

Main Methods:

• Serum samples from COPD patients and controls were analyzed using FS, RS, and SERS.
• Eight distinct machine learning algorithms were employed to analyze spectral data.
• The synthetic minority over-sampling technique (SMOTE) with gradient boosting (GB) was specifically investigated.

Main Results:

• Serum FS, RS, and SERS revealed distinct spectral variations in COPD patients compared to controls.
• Combinations of serum RS or SERS with machine learning algorithms outperformed serum FS.
• Serum SERS coupled with machine learning achieved over 0.98 accuracy for COPD vs. healthy classification.
• Serum SERS with SMOTE-GB demonstrated 0.84 accuracy for three-class classification (COPD, non-COPD, healthy).

Conclusions:

• Serum SERS combined with machine learning algorithms offers a highly accurate method for COPD detection.
• The SMOTE-GB algorithm integrated with serum SERS shows significant potential for multi-class COPD diagnosis.
• This approach represents a promising advancement in developing efficient and convenient diagnostic strategies for COPD.