Salivary ATR-FTIR Spectroscopy Coupled with Support Vector Machine Classification for Screening of Type 2 Diabetes...

Area of Science:

• Biomedical Engineering
• Spectroscopy
• Machine Learning

Background:

• Blood diagnosis of diabetes mellitus (DM) is accurate but invasive, costly, and painful.
• Alternative non-invasive methods are needed for early disease detection and management.
• ATR-FTIR spectroscopy combined with machine learning shows promise for disease screening.

Purpose of the Study:

• To develop a non-invasive diagnostic platform for type 2 DM using salivary components.
• To identify specific salivary biomarkers indicative of type 2 DM.
• To evaluate the efficacy of ATR-FTIR spectroscopy with LDA and SVM for DM diagnosis.

Main Methods:

• Utilized Attenuated Total Reflectance Fourier-Transform Infrared (ATR-FTIR) spectroscopy on saliva samples.
• Employed Linear Discriminant Analysis (LDA) and Support Vector Machine (SVM) for spectral data classification.
• Analyzed band area values at specific wavenumbers (2962, 1641, and 1073 cm⁻¹) and SHAP features.

Main Results:

• Elevated band area values at 2962 cm⁻¹, 1641 cm⁻¹, and 1073 cm⁻¹ were observed in type 2 diabetic patients.
• SVM classifier achieved the highest accuracy (87%), with 93.3% sensitivity and 74% specificity.
• SHAP analysis identified lipid and protein vibrational modes as key discriminators for DM.

Conclusions:

• ATR-FTIR spectroscopy coupled with machine learning is a viable non-invasive tool for type 2 DM screening.
• Salivary analysis offers a reagent-free, sensitive, and cost-effective alternative to traditional blood tests.
• This technology holds potential for monitoring diabetic patients and facilitating early disease detection.