Task-Dependent Synergy between FTIR and Raman Spectroscopy for Sperm Quality Prediction

Abstract Male infertility affects approximately 15% of couples worldwide, contributing to 40–50% of infertility cases in Western countries. Standard semen analysis remains the cornerstone of fertility assessment, yet conventional parameters (concentration, motility, morphology) have limited predictive value for fertilization success. Vibrational spectroscopy, particularly Fourier transform infrared (FTIR) and Raman spectroscopy, offers rapid, nondestructive chemical fingerprinting of biological samples. However, optimal strategies for integrating multimodal spectroscopic data remain unclear, and the value of fusion may depend on the biological end point. This study systematically compares FTIR and Raman spectroscopy, both independently and through multiple fusion strategies, for predicting rectilinear progressive sperm motility (binary classification) and normal morphology (continuous regression) in human seminal plasma. We determine whether complementary information exists, when fusion provides value, and which strategy optimally leverages multimodal data. Building on Raman-derived Multivariate Curve Resolution-Alternating Least Squares (MCR-ALS) concentration profiles, we further develop a clinical classifier based on four seminal plasma biomarkers (human serum albumin, glutathione S-transferase, transferrin, lysophosphatidylcholine), validated by leave-one-patient-out cross-validation and Firth’s penalized logistic regression to address near-complete separation in this small-sample setting (n = 30). Our findings reveal technique-specific advantages: Raman spectroscopy substantially outperforms FTIR for binary motility classification, with feature-level fusion matching but not exceeding Raman’s performance, indicating limited complementarity. Conversely, FTIR alone achieves superior predictive accuracy for continuous morphological outcomes. The clinical classifier based on MCR-ALS-derived biomarkers achieved AUC (Area Under the ROC Curve) = 0.855, accuracy = 86.7%, and BER (Balanced Error Rate) = 0.136, with model significance confirmed by permutation testing. Elevated human serum albumin and lysophosphatidylcholine emerged as the strongest negative predictors of high motility, consistent with seminal tract inflammation and sperm-cytotoxic phospholipid accumulation, while transferrin showed a positive association in line with its role in iron availability for sperm function. These results demonstrate that optimal spectroscopic approaches depend on the biological end point, with Raman’s structural sensitivity favoring motility assessment and FTIR’s compositional profiling better capturing morphological variation. The biomarker-based classifier is easier to interpret and complements whole-spectrum models by bridging chemometric analysis and biological interpretation.