Lactose particle size analysis using FT-NIR spectroscopy

Lactose particle size analysis using FT-NIR spectroscopy — Application Note Summary

Significance of the topic

The particle size of excipients such as lactose strongly influences powder flow, compressibility, dissolution rate, reaction kinetics and final tablet integrity. Rapid and reliable particle size assessment at-line or in-process reduces reliance on slow gravimetric sieve analyses, shortens QC turnaround, and enables better process control and product quality. FT‑NIR offers a fast, non-destructive approach that can capture both chemical and physical information from diffuse reflectance spectra, making it an attractive tool for particle size classification in pharmaceutical manufacturing.

Objectives and overview of the study

The study evaluated whether FT‑NIR diffuse reflectance spectra collected with the Thermo Scientific Antaris II MDS analyzer can discriminate between different particle size classes of pharmaceutical‑grade lactose monohydrate. The work compared raw spectral data and common preprocessing approaches (first and second derivatives) to determine which spectral format preserves particle size information most effectively for classification using chemometric methods (PCA, discriminant analysis, Mahalanobis distance).

Methodology and experimental design

• Samples: Pharmaceutical‑grade lactose monohydrate obtained in discrete mesh classes (mesh designations reported in the original note included 50, 80, 110 and 125; corresponding micrometer ranges were provided in the source table).
• Sampling: Ten replicate scans per mesh class were measured.
• Instrument and conditions: Spectra were collected on the Antaris II MDS FT‑NIR Analyzer using the Integrating Sphere (diffuse reflectance) over 4000–10000 cm⁻¹ with samples placed in a spinning sample cup.
• Data treatment: Raw absorbance spectra and first‑ and second‑derivative spectra were evaluated. Chemometric analysis used Principal Component Analysis (PCA) for visualization and patterning, discriminant analysis for classification, and Mahalanobis distance ratios to quantify class separation and potential misclassification risk (ratio of distance to nearest class vs next‑nearest class).

Použitá instrumentace

• Thermo Scientific Antaris II MDS FT‑NIR Analyzer with Integrating Sphere module and spinning sample cup.
• TQ Analyst software for discriminant analysis, PCA score plots and Mahalanobis distance calculations.

Main results and discussion

• Raw spectra: Representative raw NIR spectra showed baseline offsets and slope differences among particle size classes, consistent with particle‑size dependent scattering effects. PCA of raw spectra produced clear grouping and separation between the mesh classes, enabling successful classification.
• First derivative: First‑derivative preprocessing (which removes baseline offsets) reduced the observed class separation in PCA space but maintained useful discrimination for several classes. Mahalanobis distance ratios were smaller than for raw spectra, indicating closer proximity to neighboring classes and increased potential for misclassification compared with the raw data.
• Second derivative: Second‑derivative preprocessing (which reduces baseline slope) further degraded class separation. PCA score plots revealed particular deterioration in separation between some adjacent mesh classes (notably the 80 vs 100 mesh designations). Mahalanobis ratios showed several values <1, indicating that some samples were closer to an incorrect class than to their true class — i.e., clear misclassification risk.
• Interpretation: Physical differences due to scattering and diffuse reflectance (baseline offsets and slopes) carry the particle size signal in the NIR spectra. Derivative preprocessing commonly used to emphasize chemical features can inadvertently remove this scattering information and thereby reduce the ability to classify particle size.

Key numerical findings (summary of table/figures)

• PCA score plots: Raw spectra showed the most distinct clustering by particle size; first derivative showed reduced but still useful clustering; second derivative showed poor clustering with overlapping classes.
• Mahalanobis distance ratios: Highest (best) ratios observed for raw spectra, lower ratios for first derivative, and multiple ratios <1 for second derivative indicating misclassification. The study used these ratios to quantify how strongly samples belonged to their assigned class versus the nearest incorrect class.

Benefits and practical applications

• Rapid, non‑destructive particle size screening: FT‑NIR can provide immediate classification of particle size classes without time‑consuming sieve and gravimetric methods.
• At‑line and potentially inline monitoring: The approach supports faster QC decisions and process adjustments to maintain consistent material properties.
• Reduced operator skill requirements: Automated spectra acquisition and chemometric classification simplify routine use compared with manual sieve analysis.
• Preservation of process‑relevant information: Using raw spectra retains physical scattering markers that correlate with particle size, enabling robust classification when appropriate preprocessing is avoided.

Limitations and practical considerations

• Calibration and transferability: Models must be trained across expected material and presentation variability (batch differences, packing density, moisture) to be robust in production environments.
• Sample presentation: Integrating sphere and spinning cup were used here; different sampling accessories or packing states will affect scattering and may require model adjustment.
• Preprocessing trade‑offs: While derivatives improve chemical spectral analysis by removing baseline effects, they can remove useful physical information for particle sizing. Choice of preprocessing must be matched to the intended analytical target.

Future trends and opportunities for use

• Advanced chemometrics and machine learning: Supervised algorithms (PLS‑DA, SVM, random forests, neural networks) could improve robustness and handle more complex sample sets and overlapping classes.
• Inline and real‑time monitoring: Integration of FT‑NIR probes or in‑line sampling with continuous processing could enable true real‑time particle size surveillance and closed‑loop control.
• Hyperspectral imaging / imaging NIR: Spatially resolved NIR could map particle size distribution across a sample bed rather than classifying bulk classes.
• Calibration transfer and standardization: Development of robust transfer protocols, standard reference materials and domain adaptation will improve model portability between instruments and sites.
• Hybrid approaches: Combining NIR with complementary techniques (e.g., laser diffraction, dynamic image analysis) can provide both rapid screening and detailed PSD characterization for critical applications.

Conclusions

The application note demonstrates that diffuse reflectance FT‑NIR spectra acquired with the Antaris II can discriminate lactose particle size classes. Importantly, raw spectral data—where baseline offsets and slopes produced by scattering remain intact—delivered the best classification performance. Conventional derivative preprocessing, while valuable for chemical analysis, removed the scattering‑related signal and degraded particle size discrimination. For practical implementation, careful calibration, attention to sampling modality, and selection of preprocessing aligned with the analytical goal are essential.

References

1. Jeffrey Hirsch and Todd Strother. Lactose particle size analysis using FT‑NIR spectroscopy. Thermo Fisher Scientific Application Note AN51557 (2022).