Spectral Region Optimization and Machine Learning-Based Nonlinear Spectral Analysis for Raman Detection of Cardiac Fibrosis Following Myocardial Infarction

Cardiac fibrosis following myocardial infarction plays a critical role in the formation of scar tissue and contributes to ventricular arrhythmias, including ventricular tachycardia and sudden cardiac death. Current clinical diagnostics use electrical and structural markers, but lack precision due to low spatial resolution and absence of molecular information. In this paper, we employed line scan Raman microspectroscopy to classify sheep myocardial tissue into muscle, necrotic, granulated, and fibrotic tissue types, using collagen as a molecular biomarker. Three spectral regions were evaluated: region A (600-2960 cm^-1), region B (600-1399 cm^-1 and 1751-2960 cm^-1), and region C (1400-1750 cm^-1), which includes the prominent collagen-associated peaks at 1448 cm^-1 and 1652 cm^-1. Linear and nonlinear principal component analysis (PCA) and support vector machines (SVMs) were applied for dimensionality reduction and classification, with nonlinear models specifically addressing the nonlinearity of collagen formation during fibrogenesis. Histological validation was performed using Masson's trichrome staining. Raman bands associated with collagen in region C consistently outperformed regions A and B, achieving the highest explained variance and best class separation in both binary and multiclass PCA models for both linear and nonlinear approaches. The ratio of collagen-related peaks enabled stage-dependent tissue characterization, confirming the nonlinear nature of fibrotic remodeling. Our findings highlight the diagnostic potential of collagen-associated Raman bands for characterizing myocardial fibrosis. The proposed PCA-SVM framework demonstrates robust performance even with limited sample size and has the potential to lay the foundation for real-time intraoperative diagnostics.