Label-Free Biochemical Characterization and Subtype Classification of Human Breast Pathologies Using Fingerprint Raman Spectroscopy and Machine Learning

Breast cancer remains a major cause of cancer-related mortality, and improved molecular characterization of pathological progression may support more objective tissue assessment. Raman spectroscopy (RS) provides a label-free and non-destructive approach for probing the intrinsic biochemical composition of biological tissues. In this study, we establish a Raman-based framework for characterizing biochemical remodeling across representative human breast pathologies and for exploring subtype-associated spectral differences between invasive ductal carcinoma (IDC) and invasive lobular carcinoma (ILC). Formalin-fixed, paraffin-embedded (FFPE) tissues from five patients, representing normal breast tissue, intraductal papilloma, high-grade ductal carcinoma in situ (DCIS), IDC, and ILC, were analyzed using 785 nm excitation over the fingerprint region of 500–2000 cm⁻¹. Spectral preprocessing combining Savitzky–Golay smoothing, airPLS baseline correction, and vector normalization substantially reduced fluorescence-related background contributions and improved the resolution of biochemical Raman features. The spectra revealed progressive attenuation of lipid-associated bands near ~1063, ~1300, and ~1445 cm⁻¹, accompanied by increased relative contributions from protein- and nucleic-acid-associated features near ~1003, ~1420, and 1600–1650 cm⁻¹. The phenylalanine-to-lipid intensity ratio (I₁₀₀₃/I₁₄₄₅) increased from ≤0.2 in normal tissue to ≥1.5 in invasive lesions, providing an interpretable quantitative descriptor of the shift from a lipid-dominant toward a relatively protein-enriched Raman phenotype. IDC and ILC also exhibited distinct spectral characteristics, with IDC showing a prominent ~1618 cm⁻¹ aromatic C=C-associated feature and ILC exhibiting an enhanced ~890 cm⁻¹ carbohydrate-associated contribution. Principal component analysis (PCA) demonstrated clear multivariate organization of the training spectra, while supervised classification using LDA, SVM, random forest, and KNN confirmed that the Raman-derived features contained substantial discriminative information. Among the evaluated models, the RBF-SVM achieved the highest training accuracy of 93.4% and retained an accuracy of 90.0%, sensitivity of 85.7%, and specificity of 100% on a held-out test set. Collectively, these findings support a proof-of-concept framework in which breast pathological progression is associated with coordinated biochemical Raman changes that can be quantified through interpretable spectral descriptors and subsequently exploited for machine-learning classification. Larger patient-level and multi-center studies are required to validate the robustness and clinical generalizability of the proposed approach.