AI-POWERED ANALYSIS

Whole blood samples were analyzed using FTIR spectroscopy, capturing molecular vibrations indicative of biochemical components. Data pre-processing involved min-max normalization and baseline correction. Principal Component Analysis (PCA) was first applied for unsupervised dimensionality reduction and to identify inherent clustering. Subsequently, supervised machine learning models—Decision Tree (DT), Random Forest (RF), k-Nearest Neighbor (k-NN), Support Vector Machine (SVM), and Neural Network (NN)—were trained and evaluated using cross-validation (LOGO CV), with hyperparameter optimization for optimal diagnostic and classification potential in SSc. The analysis focused on the mid-infrared fingerprint region (800-1800 cm⁻¹).

FTIR spectra revealed subtle but consistent differences, particularly in amide I/II and lipid-associated regions, correlating with known metabolomic alterations in SSc. PCA demonstrated clear clustering along the first principal component, explaining over 84% of variance and highlighting key wavenumber regions (1000-1200 cm⁻¹ and 1500-1700 cm⁻¹) related to proteins, lipids, and carbohydrates. For SSc subtype classification, the Random Forest model achieved the highest accuracy of 81.03% after optimization, with a positive MCC, indicating robust performance. While ILD classification showed potential, its accuracy was moderate (Table 4, Neural Network at 63.22% CA).

The findings suggest that FTIR spectroscopy, especially when integrated with machine learning, holds significant promise as a rapid, non-invasive tool for disease stratification and biomarker discovery in SSc. This approach could complement existing diagnostic methods, offering more precise clinical subtyping and facilitating personalized treatment strategies. Further research is needed to refine spectral pre-processing, optimize models, expand patient cohorts, and integrate with other omics techniques to enhance clinical applicability and correlate findings with disease progression and treatment response.