| Sigma Xi Outstanding Poster Award - 2006 FDA Science Forum |
M. L. Hamad, C. D. Ellison, A. S. Carlin, E. H. Jefferson, M. A. Khan, R. C. Lyon, DPQR, CDER, FDA, Silver Spring, MD
Background: Spectroscopic and imaging techniques for chemical and physical analysis of pharmaceutical products and processes offer significant advantages over traditional analytical methods. However, representative sampling of heterogeneous pharmaceutical materials and products during process monitoring is challenging. It is necessary to determine the appropriate spot size to measure a representative sample. Intuitively, an image is homogeneous at a given resolution if most pixels are representative of the whole. This study explores algorithms that produce quantitative descriptions of heterogeneity yielding better sampling.
Methods:Near-infrared chemical images were acquired of acetaminophen tablets of varying strengths ranging from 60-120 mg. Chemometrics was applied to each NIR image to create a concentration image for acetaminophen.Images were 256x320 pixels. "Macropixels" were constructed by binning together neighboring pixels, representing sub-samples of the original image.Histograms were used to express the distributions of pixel and macropixel intensity values. Different metrics were evaluated for their ability to characterize image homogeneity.
Results: Digital images were analyzed using histogram statistics and co-occurrence matrix analysis. In all cases, homogeneity improved as the macropixel size increased. The minimum appropriate spot size for representative sampling was predicted.
Conclusions: Algorithms for quantitatively evaluating the level of heterogeneity in chemical images were evaluated. These algorithms will serve as useful tools when examining the spatial characteristics of drug products and pharmaceutical blends with chemical imaging. This method of analysis can also be used to determine the sampling requirements of heterogeneous materials undergoing analysis with non-spectroscopic techniques.