Lactose particle size analysis using FT-NIR spectroscopy

Significance of the topic

Particle size of pharmaceutical powders strongly influences flowability, compressibility, dissolution rate, reaction kinetics and final dosage form quality. Routine confirmation of raw-material particle size helps optimize manufacturing, reduce batch failures and improve product performance. Near-infrared (NIR) spectroscopy offers a rapid, non-destructive alternative to traditional sieve/gravimetric methods, enabling faster decision-making and potential at-line or in-line monitoring in pharmaceutical production.

Objectives and overview of the study

The application note evaluated the ability of FT-NIR spectroscopy (Thermo Scientific Antaris II MDS) combined with multivariate analysis to discriminate lactose monohydrate powders of different particle-size classes (mesh sizes 50–125 µm). The study compared spectral preprocessing approaches (raw spectra, first derivative, second derivative) for classification performance using Principal Component Analysis (PCA), Discriminant Analysis and Mahalanobis-distance metrics implemented in TQ Analyst software.

Methodology

• Samples: Pharmaceutical-grade lactose monohydrate (Pharmatose), mesh sizes spanning approximately 50–125 µm; ten replicates per mesh size.
• Data acquisition: Diffuse-reflectance FT-NIR spectra collected on Antaris II MDS with Integrating Sphere and spinning sample cup; spectral range 4000–10000 cm⁻¹.
• Data processing and chemometrics: Discriminant Analysis for classification, PCA for visualization of spectral variance, and Mahalanobis distance ratios to quantify class separation. Spectral formats tested: raw absorbance, first derivative (to remove baseline offsets), and second derivative (to remove baseline slope).

Used instrumentation

• Thermo Scientific Antaris II MDS FT-NIR Analyzer
• Integrating sphere diffuse-reflectance module with spinning sample cup holder
• TQ Analyst software for Discriminant Analysis, PCA and Mahalanobis-distance calculations

Main results and discussion

• Raw spectra contained baseline offsets and slope differences that correlated with particle-size classes. PCA score plots for raw spectra showed clear grouping and separation of different mesh sizes, indicating that particle-size information is encoded in baseline/offset features.
• First-derivative preprocessing reduced baseline offsets and preserved acceptable but reduced class separation in PCA space; discrimination performance decreased compared with raw spectra.
• Second-derivative preprocessing, which removes baseline slope, further degraded class separation; some classes (notably 80 and 100 mesh) became poorly separated and exhibited misclassification.
• Mahalanobis-distance ratio analysis quantified these observations: raw spectra produced large ratios (good separation), while derivative-processed spectra showed smaller ratios and several instances where samples were closer to an incorrect class (ratios <1 indicating misclassification).
• Conclusion from data: physical differences related to particle size manifest primarily as baseline offsets/slopes in NIR diffuse-reflectance spectra; thus, standard derivative preprocessing commonly used to emphasize chemical features can remove the relevant physical signal and impair classification.

Benefits and practical applications of the method

• Rapid, non-destructive determination of particle-size classes without the need for sieving and gravimetric analysis; suitable for at-line or near-line quality checks.
• Reduces time to result and dependence on specialized operators, improving manufacturing throughput and enabling faster corrective actions.
• Applicability as a PAT (process analytical technology) tool for monitoring raw-material variability and ensuring consistent feedstock properties to downstream processes (e.g., tableting, coating).
• Enables multivariate classification workflows to be integrated into QC/production systems for automated pass/fail or binning decisions.

Limitations and considerations

• Observations are material-specific: calibration and model performance will depend on the excipient, moisture content, packing density and optical properties of the powder.
• Representative sampling and sample presentation (spinning cup, integrating sphere) are critical to minimize orientation and packing artifacts.
• Spectral preprocessing must be chosen to preserve the physical information of interest; defaults optimized for chemical analysis (derivatives) may be inappropriate for particle-size classification.
• Quantitative particle-size distribution estimation (vs. class discrimination) would require robust calibration across broader particle-size ranges and formulations.

Future trends and potential applications

• Development of regression models (e.g., multivariate partial least squares) to predict continuous particle-size parameters or distributions rather than discrete classes.
• Integration of FT-NIR sensors into continuous manufacturing lines for real-time particle-size monitoring and feedback control.
• Combining NIR with complementary techniques (laser diffraction, imaging, Raman) or advanced machine-learning algorithms to improve robustness across materials and formulations.
• Improvements in calibration transfer and standardization to enable method portability between instruments and sites.
• Exploration of scattering-sensitive preprocessing or physically informed models that explicitly account for particle-scattering effects rather than removing them.

Conclusion

FT-NIR diffuse-reflectance spectroscopy using the Antaris II MDS system can reliably discriminate lactose particle-size classes when raw spectral information is used. Standard derivative preprocessing that removes baseline offsets and slopes tends to eliminate the spectral signatures associated with particle size and reduces classification performance. The method offers a fast, non-destructive alternative to traditional sieve/gravimetric testing with clear potential for PAT implementation, provided calibrations are tailored to the material and sampling protocol is well controlled.