(166g) Multivariate Texture and Image Analysis for Solid Dosage Intermediate and Final Product Characterization and Downstream Performance Prediction

Authors: 
Garcia-Munoz, S., Pfizer Global Research & Development
Carmody, A., Worldwide Research and Development, Pfizer Inc.


A framework using wavelet decomposition coupled with multivariate latent variable methods is proposed to quantitatively characterize the texture appearance of pharmaceutical materials and products. This framework allows having a non-biased measure of the visual characteristic of solid materials. The proposed technique has shown successful application in the steel sector , and is applied in a straight forward manner to grayscale images of the pharmaceutical products (or intermediate materials).

In summary, the technique uses wavelet decomposition on the grayscale images to reduce them to a vector of energies for multiple ?frequencies? (called details in wavelet theory), such vectors are then compressed using Principal Components Analysis (for characterization) or used as regressors against a property of interest (e.g. in vitro dissolution, compression profile)

This method is illustrated with two case studies:

i) The analysis of an intermediate pharmaceutical product (wet granules from a high shear wet granulation step) to provide a quantitative envelope for the material, and also to predict its manufacturability at the compression step (after drying and milling).

ii) The characterization of a modified release film coated tablets, where the textural characteristics of the tablet are linked to the in vitro dissolution of the product.

Using a latent variable regression model, it was found that the appearance of the wet granules is able to predict 80% of the hardness compression profile for the material. In the case of the modified release tablets, the visual characteristics of the tablet alone can predict dissolution at 9hr with a 91% R2 accuracy. A model that considers the processing conditions in the film coater and the visual appearance is able to predict dissolution to a 97% R2.