The aim of this study was to assess the potential of FTIR spectroscopy for monitoring biochemical changes in serum samples of individuals with carotid atherosclerosis following surgical intervention. Principal Component Analysis (PCA) of FTIR spectra from serum samples reveals distinct biochemical patterns at different time points: pre-surgery, 24 h post-surgery, and 48 h post-surgery. Two spectral ranges, 800–1800 cm−1 and 2800–3000 cm−1, were analyzed. PCA demonstrated that pre-surgery samples can be clearly differentiated from those taken 24 and 48 h post-surgery. However, no significant distinction was found between the 24-hour and 48-hour post-surgery samples. For the 800–1800 cm−1 range, the first principal component (PC1) explained 77.49% of the variance, highlighting the molecular vibrations of lipids, proteins, and carbohydrates. In the 2800–3000 cm−1 range, PC1 accounted for 94.89% of the variance, primarily reflecting lipid-related vibrations. These findings indicate a clear separation between pre-surgery and post-surgery samples, with the most significant variance explained by PC1. Additionally, the Boruta algorithm identified a key spectral range between 1506 cm−1 and 1673 cm−1, critical for distinguishing the samples. Classification models, including k-Nearest Neighbors, Gradient Boosting, Support Vector Machine, and Neural Network, demonstrated excellent performance in differentiating pre-surgery and post-surgery samples. However, the models struggled to distinguish between the 24-hour and 48-hour post-surgery time points. This suggests that FTIR spectroscopy may be useful for monitoring post-surgery recovery in carotid artery atherosclerosis, although subtle changes in the biochemical profile are challenging to detect between 24 and 48 h post-surgery.
Attenuated total reflectance-Fourier transform infrared spectroscopy (ATR-FTIR) provides information on the molecular composition and structure of samples. The use of ATR-FTIR was evaluated for biochemical analysis and taxonomic differentiation of entomopathogenic nematodes (EPNs). Spectra were obtained from a small sample (pellet) of a nematode population recovered from commercial EPN packages, which was placed directly on the ATR plate. Differences in signal intensity at multiple peaks associated with biomolecules critical to the survival of EPN (trehalose, glycogen, and triglyceride) were measured and visualized using Non-Metric Multidimensional Scaling (nMDS) and Principal Component Analysis (PCA). Statistically significant differences in peak signal intensity were observed between EPN species for each biochemical parameter, providing a basis for assessing the likelihood of their performance success in the field conditions. The present study also evaluated FTIR analysis of EPN for taxonomic differentiation. Results demonstrate that FTIR can be used to identify and differentiate Steinernema and Heterorhabditis genera/species, offering a potentially faster, less expensive alternative to molecular identification techniques. Ultimately, this study demonstrates the efficacy of ATR-FTIR as a reliable method for assessing the biochemical suitability of EPN products for field applications and differentiating between EPNs.
Endometrial cancer (EC) is increasingly prevalent worldwide, highlighting the need for non-invasive blood-based diagnostic triage tools. ATR-FTIR spectroscopy enables rapid, label-free biochemical profiling of plasma or serum for experimental cancer detection. To date, no systematic review or meta-analysis has evaluated the experimental performance of infrared spectroscopy for discriminating EC from non-cancer in blood-based samples. This study synthesizes available evidence to characterize the strength, consistency, and heterogeneity of the underlying spectroscopic signal across preclinical and proof-of-concept studies. MEDLINE, Web of Science, EMBASE, Scopus, Google Scholar, and CENTRAL were searched without language restrictions. Eligible studies evaluated ATR-FTIR spectroscopy of plasma or serum using histopathology as the reference standard. Pooled sensitivity, specificity, likelihood ratios, and diagnostic odds ratios were estimated using a bivariate random-effects model, with assessment of heterogeneity, threshold effects, and publication bias. Five case–control studies comprising 1376 participants were included. For plasma-based analyses, pooled sensitivity was 0.61 (95% CI: 0.59–0.68) and specificity was 0.73 (95% CI: 0.69–0.76), with a diagnostic odds ratio of 4.23 (95% CI: 3.33–5.37). For serum-based analyses, pooled sensitivity and specificity were both 0.62 (95% CI: 0.59–0.65), with a diagnostic odds ratio of 2.65 (95% CI: 2.16–3.25). Substantial heterogeneity and significant threshold effects were observed. Current evidence supports reproducible spectroscopic differences between EC and non-cancer blood samples under experimental conditions. However, methodological heterogeneity and retrospective case–control study designs limit clinical interpretability. These findings provide a benchmark for future prospective validation rather than immediate clinical application.
Serum attenuated total reflection Fourier-transform infrared (ATR-FTIR) spectroscopy has been widely investigated for cancer-associated biofluid classification, but the analytical robustness of such models under acquisition-batch variation, atmosphere-sensitive spectral intervals, and correlated wavenumber selection remains insufficiently evaluated. Here, serum ATR-FTIR fingerprints from 306 participants, including lung cancer (LC, n = 144), benign lung disease (BEN, n = 98), and healthy controls (HC, n = 64), were analyzed using a robustness-oriented chemometric workflow. The primary LC/BEN/HC task was evaluated by leave-one-BoxID-group-out validation across true acquisition groups, with all preprocessing, normalization, feature selection, and model training confined to the training data within each validation split. To assess whether classification depended on background-sensitive spectral information, the 2400-2700cm⁻¹ atmosphere-sensitive region was further masked and compared with the full-spectrum setting. Elastic Net feature selection combined with ridge logistic ECOC achieved the best tri-class performance, yielding accuracies of 0.992 ± 0.013 for the full-spectrum model and 0.985 ± 0.021 after masking. Label-permutation analysis of the primary MASKED model yielded a null Macro F1 of 0.281 ± 0.034, far below the observed Macro F1 of 0.986, supporting non-random tri-class discrimination. Stable discriminative features were mainly located in the amide bands, fingerprint region, and selected high-wavenumber regions, supporting band-level spectrochemical interpretation rather than isolated diagnostic wavenumbers. An additional exploratory stage-associated analysis using repeated stratified five-fold cross-validation with 10 repeats indicated internally validated spectral discrimination within the LC cohort, although independent validation in larger, stage-balanced cohorts is still required. Together, these results support a robustness-oriented ATR-FTIR chemometric workflow for lung-cancer-associated serum discrimination, emphasizing acquisition-group validation, atmospheric-region masking, and band-level spectral interpretation.
Surface-enhanced Raman spectroscopy (SERS) enables rapid, label-free bacterial fingerprinting, yet many machine-learning studies lack transparent evaluation protocols and interpretable feature analysis. This study presents a SERS-machine-learning evaluation framework for bacterial classification using a fixed-concentration ten-class Staphylococcus aureus dataset (100 spectra) and leak-free 60/20/20 train/validation/test partitions repeated across 15 random seeds. After Savitzky-Golay smoothing and standard normal variate normalization, spectra were analyzed in both full (384 variables) and cropped (400–1200 cm−1; 298 variables) representations. Linear discriminant analysis (LDA) served as the primary dense classifier, while stability-selection-based feature selection combined with LDA (StabSel-LDA) was evaluated as a sparse companion model with validation-tuned feature count K. LDA achieved near-ceiling clean performance (mean accuracy 0.970–0.973; macro-F1 0.966–0.969) and remained stable under additive noise, drift, and ±5 cm−1 spectral shifts (accuracy 0.953). StabSel-LDA reduced the feature set to 124–228 bands and produced reproducible band-selection profiles across seeds, but showed lower clean accuracy (0.940–0.943) and pronounced degradation under ±5 cm−1 shifts (0.760–0.783). These results support a role-separated deployment strategy in which LDA functions as the primary decision model, while StabSel-LDA provides a sparse companion for feature compression and interpretable spectral-band identification. The proposed framework offers a transparent approach for balancing classification performance, robustness assessment, and spectral interpretability in small SERS datasets.
Metabolic syndrome (MetS) is characterized by central adiposity, hypertension, dyslipidemia, insulin resistance, and chronic inflammation, conditions that alter lipid regulation in multiple tissues, including skin. Fatty acidbinding proteins, particularly epidermal fatty acid-binding protein (E-FABP, FABP5), have been proposed as potential biomarkers of these metabolic and cardiovascular alterations; however, their contribution in skin has not been studied using vibrational spectroscopy. This work aimed to evaluate whether Raman spectroscopy can detect epidermal FABP-associated molecular changes in skin during the development of MetS. Male Wistar rats were fed a high-fat diet for up to 52 weeks, and in vivo Raman spectra of abdominal skin were acquired at several time points. Raman band-intensity analysis, principal component analysis, and logistic regression classification were applied to identify biochemical differences between the control and metabolic syndrome groups. The Raman bands at 1002 cm 1 , 1050 cm-1 , and 1377 cm 1 , associated with E-FABP, showed time-dependent differences, with the most significant increase at 18 weeks and convergence between groups by 52 weeks. Based on this unique spectral feature, we developed a noninvasive Raman workflow capable of detecting E-FAPB directly in living tissue, as a potential Raman biomarker of MetS. The classification model achieved 79% accuracy with an area under the curve of 0.85. In summary, our findings indicate that Raman spectroscopy can detect E-FABP alterations in skin associated with MetS, supporting its potential as a non-invasive tool for early metabolic assessment.
Aqueous thin layers on germanium surface have been investigated by infrared spectroscopy. Two doublets of water librations and one of atmospheric CO2 bound in the thin layer imply the aqueous thin layer structuring, namely formation of quasi-crystal structure. The layer structuring and theoretical estimations based on polaronic exciton concept allow to assign a broad band at 1124 cm-1 to longitudinal optical phonons observed in the IR spectrum. The building up of the broad band is interpreted in terms of Fermi resonance that involves two frequencies at 1556 and 1664 cm-1. Fermi coupling coefficient with a slight (36 cm-1) or strong perturbation of 220 cm-1 partakes in the mixing with the energy levels. The mixing results in unusual features of the IR spectra of the thin aqueous layers observed on germanium plates. It seems that the appearance of Fermi resonance in the structure of aqueous layer arises because of a brake of spin-orbit coupling in polaronic exciton. Namely, it arises because the repulsion between the spins of the opposite charges.\
To achieve accurate quality assessment of B. dorsalis-infested apples, this study proposes an early detection and firmness prediction method based on hyperspectral imaging (HSI) integrated with a one-dimensional convolutional neural network (1D-CNN). First, a hyperspectral imaging system (400-1000 nm) was used to acquire images over two consecutive years for three categories of apple samples: healthy, B. dorsalis-infested, and superficially similar scarred apples, while the firmness of each fruit was measured simultaneously. Subsequently, the successive projections algorithm (SPA) and competitive adaptive reweighted sampling (CARS) were applied to select characteristic bands related to fruit categories and firmness, respectively. Finally, early detection and firmness prediction models were developed by integrating partial least squares-discriminant analysis (PLS-DA), back-propagation neural network (BP), 1D-CNN, random forest (RF), support vector machine (SVM), and 1DCNN. The results showed that the SPA-1D-CNN detection model with 10 selected characteristic bands achieved satisfactory performance, with an overall classification accuracy of 94%, precision of 0.94, recall of 0.94, and an F1-score of 0.94. For firmness prediction, the SPA-1D-CNN model also performed well, For the calibration set, R2 c was 0.87, RMSEC was 0.76. For the prediction set, R2p was 0.86, RMSEP was 0.79. Moreover, pixel-wise firmness prediction was implemented on the spectral images, enabling the visualization of apple firmness distribution. This study demonstrates that HSI combined with 1D-CNN provides an effective approach for early quality assessment of B. dorsalis-infested apples, providing theoretical support for the development of relevant detection devices.
The optical path length of a transmission flow cell is a critical parameter that influences the quantitative accuracy of infrared (IR) spectroscopy. This study investigates how path length affects the accuracy of IR spectroscopy for measuring bulk properties using a generalizable machine learning (ML)-based framework. The approach was demonstrated using an onboard fuel sensing application aimed at predicting the derived cetane number (DCN) of jet fuels and fuel blends from mid-IR spectra in the wavelength range of 5.8-7.2 & micro;m. A range of path lengths, from 1 & times; 10-5 to 500 & micro;m, was examined by simulating spectra through scaling of experimentally measured data. The results showed that extremely short path lengths led to poor prediction accuracy due to insufficient absorption and low signal-to-noise ratio (SNR), while excessively long path lengths caused strong absorption saturation and amplified spectral artifacts, degrading model performance. A broad near-optimal range between 10 and 58.7 & micro;m was identified within the simulation framework, where spectra exhibited adequate absorption, moderate saturation, and minimal distortion. These findings demonstrate that allowing moderate absorption saturation can enhance predictive accuracy by improving SNR at wavelengths most correlated with the target property and emphasize that optimal measurement conditions should be determined based on predictive performance rather than solely maximizing SNR while maintaining absorbance within traditional limits.
The growing electronic cigarette market faces significant quality control issues regarding e-liquid compositions. Fast analytical methods are required to verify the proportions of the primary solvents, vegetable glycerin and propylene glycol. This study used Attenuated Total Reflectance Fourier Transform Infrared spectroscopy with partial least squares regression to quantify glycerin and propylene glycol in commercial e-liquids. Gas Chromatography with Flame Ionization Detection served as the reference method. A matrix-matched calibration model, built with real commercial samples, was developed to effectively account for spectral interferences. The optimized PLS model achieved high predictive accuracy (R-2 = 0.9995), strongly correlating with the chromatography results. Crucially, both techniques exposed severe discrepancies between the measured glycerin/propylene glycol ratios and the manufacturers declared labels. These findings validate this spectroscopic approach as a robust, non-destructive, and low-cost tool for the routine quality control and rapid screening of mislabeled formulations. Future work should expand the calibration dataset with a larger set of commercial samples to further enhance the model's universal applicability.
Closed-form solutions of isolated spheres have been used extensively in modeling and understanding spherical biological scattering systems in infrared microscopy. However, since samples are deposited onto infrared transparent slides, they are not isolated, and the slide may affect the measured spectra. While the effect of the slides used in infrared microscopy is well understood for thin film samples, it is not understood to what extent the slide affects the scattering signatures from spherical systems. To get a more accurate representation of the measurements, the sample and the slide should be considered as a combined scattering system. In the present study, we investigate the effect of microscope slides of different materials on the measured absorbance spectra. The primary effect is dampening of the ripples when the slide is index matched to the spherical sample, while the wiggles are not affected by the slide. Since spectra of spherical biological samples most often contain wiggles but are devoid of ripples, the effect of the slide is in many cases not significant. Since forward models in use for spectral correction only consider the wiggles, one can conclude that the forward models are valid regardless of the slide used in the experiment.
To address the issues of non-real-time data acquisition and sample consistency in liquid-phase near-infrared spectroscopy, this study designed and developed an online liquid-phase near-infrared spectroscopy acquisition system based on an STM32 microcontroller. The system innovatively employs a dual-buffer bottle liquid-level closed-loop control strategy, aiming to eliminate pipeline cavitation at the hardware level and enhance sample delivery efficiency. Comparative experiments using distilled water samples demonstrated that the dual-buffer bottle liquid level control system achieved 0% cavitation occurrence, a relative standard deviation (RSD) of spectral repeatability as low as 1.2%, and a 14.5% improvement in sample delivery efficiency, validating the physical stability of the hardware system. Building upon this platform, systematic data quality evaluation was conducted. Spectral data were collected from ammonia nitrogen solutions and beer samples. Competitive Adaptive Re-weighted Sampling algorithm was employed to identify key feature wavelengths for each sample type. Multiple Extreme Learning Machines regression models were established to assess the usability of chemical information within the data. Results showed that the ammonia nitrogen optimal model achieved a coefficient of determination exceeding 0.99 for both calibration and validation sets, with a relative root mean square error below 4.8%. Similarly, the ethanol optimal model achieved a coefficient of determination exceeding 0.99 for both sets, with a relative root mean square error below 1.75%. These results validate the effectiveness of the dual-buffer-bottle liquid level closed-loop control strategy within this system and confirm the usability of the spectral data collected by the system.
Traumatic brain injury (TBI) is a common case type in forensic medicine. Accurately distinguishing injured from non-injured tissues and estimating the time elapsed since injury are crucial aspects of forensic practice. This study aimed to investigate the feasibility of using Attenuated Total Reflection Fourier Transform Infrared (ATR-FTIR) spectroscopy to discriminate between injured and non-injured tissues, as well as to differentiate various post-injury intervals following moderate TBI. A mouse model of moderate TBI was established using the Feeney weight-drop method, and individuals with moderate injury were strictly selected based on the modified Neurological Severity Score (mNSS). Firstly, Principal Component Analysis (PCA) was employed for dimensionality reduction and visualization of the spectral data. Secondly, classification models including Partial Least Squares Discriminant Analysis (PLS-DA) and Support Vector Machine (SVM) were constructed to distinguish injured from non-injured tissues across different time points, both of which demonstrated excellent classification performance. Subsequently, regression models, including Partial Least Squares Regression (PLS-R), Constrained Linear Regression (CLR), Principal Component Regression (PCR), Multiple Linear Regression (MLR), Support Vector Regression (SVR), and Artificial Neural Network (ANN), were developed based on the spectral data to predict the survival time post-TBI. Among these, the SVR model delivered the best prediction performance with R2 of 0.936 and RMSE of 1.485 days. This preliminary study demonstrates that FTIR spectroscopy combined with chemometrics enables rapid and accurate identification of trauma occurrence and estimation of the post-injury interval, thereby providing a scientific basis for case investigation and judicial proceedings.
Gunshot residue (GSR) is a key form of trace evidence in firearm-related investigations It comprises of two main components namely inorganic GSR (IGSR) and organic GSR (OGSR). IGSR is traditionally classified based on particle morphology and elemental composition. However, this approach becomes challenging for residues from non-traditional ammunition, including heavy metal-free primers. Consequently, there is increasing interest in incorporating OGSR into forensic protocols and developing standardized approaches for its interpretation. As indicated in several studies, OGSR detection can be affected by environmental variables; however, a systematic study on its stability under defined temperature and humidity conditions remains limited. In this study, we evaluated the chemical stability of OGSR under controlled laboratory simulations of temperature and relative humidity. The conditions were selected to represent ambient scenarios relevant to typical Indian weather, with temperatures ranging from 10 to 60 degrees C, relative humidity from 45% to 95% RH and exposure time varying between 0 and 8 h. A total of 51 OGSR samples were collected under controlled indoor firing-range conditions using 9 & times; 19 mm ammunition. These samples were subjected to artificial ageing in environmental chambers and exposed for different durations within this time window to study the degradation pathway of OGSR. To interrogate these potential degradation pathways, OGSR stability was monitored using Raman spectroscopy to detect time-dependent vibrational spectral changes. The results demonstrate no measurable changes in OGSR profiles under the tested temperature-humidity-exposure time combinations and, highlighting factors that influence OGSR detectability when evidence collection is delayed within hours after discharge. These findings provide condition-specific data to support interpretation of OGSR in forensic casework, while noting that longer-term persistence and complex outdoor scene effects were not evaluated in the present study.
Coherent Anti-Stokes Raman Scattering (CARS) spectroscopy enables label-free chemical imaging but is strongly affected by non-resonant background (NRB) and measurement noise, which complicate Raman signal retrieval. We present TCN-BiGRU, a hybrid deep learning architecture that combines a Temporal Convolutional Network (TCN) with Bidirectional Gated Recurrent Unit (BiGRU) for long-range spectral context. The proposed model is trained using simulated CARS data with NRB and controlled noise added to emulate challenging experimental conditions, with the resonant-to-NRB ratio R sampled during training. For performance evaluation, we sweep R from 1.0 down to 0.01 and test under severe noise contamination, including Gaussian noise (standard deviation sigma up to 0.1) and low-count Poisson noise (lambda down to 102). These stress-test conditions show that the model preserves peak positions and relative intensities while suppressing the NRB to produce near-flat baselines. Compared with existing models such as SpecNet and BiLSTM, TCN-BiGRU achieves lower mean squared error (MSE) and detects Raman peaks more reliably across different R and noise settings, keeping peak locations and relative heights closer to the ground-truth resonant Raman spectrum while reducing noise-induced distortions. These results suggest that TCN-BiGRU is a promising approach for simplifying BCARS analysis in NRB-dominant, lowSNR conditions, providing a single-step route to Raman spectra without a separately measured NRB spectrum.
Chronic Obstructive Pulmonary Disease (COPD) is a widespread, progressive respiratory condition marked by ongoing airflow limitation, significantly impacting global public health and socioeconomic progress. Currently, diagnosing COPD is challenging due to early detection difficulties and lack of specificity. However, the serum Fourier infrared spectroscopy (FTIR) spectrum offers a quick method that captures metabolic spectral features, presenting a new diagnostic approach. This study evaluates the potential of diagnosing COPD using serum FTIR spectra combined with principal component analysis-linear discriminant analysis (PCA-LDA), support vector machine linear (SVM-linear), SVM- radial basis function (SVM-RBF), K-nearest neighbors (KNN) and decision tree (DT). Differences were observed at the wavelengths 1078, 1172, 1240, 1310, 1396, 1452, 1538, 1639, 2873, 2928, 2958, and 3278 cm-1 in the serum FTIR spectra of the COPD group compared to healthy individuals. By utilizing the PCA-LDA algorithm within the 2800-3080 cm-1 range, the classification accuracy was maximized. When paired with serum FTIR, this method achieved 100% accuracy, indicating promising potential for the use of serum FTIR spectroscopy and PCA-LDA in COPD detection.
This study investigates methane pyrolysis, both with and without radio-frequency plasma, to simultaneously synthesize high-value nanostructured solid carbon (s-C) materials on Cu, Ni, Fe, and Ga catalysts via chemical vapor deposition (CVD). Raman spectroscopy was employed to assess the optimum growth conditions including the influence of hydrogen during the CVD and plasma-enhanced CVD (PECVD) processes. With immediate characterization prior to transfer, mono-layer graphene grown on copper was readily identified by the presence of a split G-band (G and D '), reflecting its conformal adherence to the substrate and structural defects. In contrast, the D' - band was not present with the few or multi-layer graphene grown on nickel. This difference comes in part due to the relatively large crystalline domain size associated with well-stacked multilayer graphene and the relatively low defect density in the graphene layers. This observation was further supported by the 2D-band profile, which exhibited a sharp single peak for mono-layer graphene, in contrast to the broader, multicomponent peak characteristic of multi-layer graphene. The abundance ratio of the semiconducting and metallic single-walled carbon nanotubes grown using Fe catalysts was inferred comparing the intensity ratio of the G+ and G- bands. When Ga was used as a low-melting-point catalyst, PECVD at relatively low temperatures yielded a s-C product with a G-to-D band intensity ratio of approximately 1, indicative of carbon microtube formation. Overall, the effects of the catalyst, plasma, temperature, pressure, and the gas environment on s-C growth were investigated using Raman spectroscopy.
Resazurin is a widely used redox-active dye whose reduction to resorufin forms the basis of numerous biological viability and metabolic assays. Despite its extensive use, a detailed vibrational spectroscopic characterization of the resazurin-resorufin redox transformation, supported by rigorous molecular-level band assignments, particularly in biologically relevant environments, remains lacking. Here, we investigate vibrational spectroscopic monitoring of the resazurin-resorufin redox transformation during bacterial metabolic activity in an Escherichia coli strain, using a correlative approach that integrates Raman spectroscopy, infrared (IR) spectroscopy, polarization-dependent Raman measurements, and density functional theory (DFT) simulations. Raman and IR spectra collected during the redox process show clear and reproducible changes in both band positions and relative intensities. In particular, a consistent Raman band shift from 1643 to 1652 cm-1 is observed, along with pronounced intensity redistribution in low-frequency modes associated with ring deformation and aromatic skeletal vibrations. Quantitative analysis of selected Raman band ratios in the 480-450, 1179-1149, and 1450-1397 cm-1 regions provides a simple and robust way to follow the progression of the redox transformation. Polarization-dependent Raman measurements further reveal changes in vibrational symmetry and molecular polarizability as reduction proceeds. DFT simulations reproduce the key experimental features and enable assignment of the observed Raman bands to specific molecular motions. UV-Visible and fluorescence measurements provide independent confirmation of the redox state of the dye. Together, this work establishes a DFTsupported vibrational framework for monitoring the resazurin-resorufin redox transformation in biological systems, providing a molecular-level basis for future spectroscopic studies employing resazurin as a redox reporter.
