A new criterion to assess distributional homogeneity in hyperspectral images of solid pharmaceutical dosage forms

During galenic formulation development, homogeneity of distribution is a critical parameter to check since it may influence activity and safety of the drug. Several techniques exist to assess this homogeneity, the most used and recognized being HPLC. However, these techniques are destructive, time consuming and uses a lot of organic solvents. Vibrational spectroscopies are promising green chemistry techniques that may replace HPLC for several analysis tasks thanks of their rapid, non-destructive and non-pollutant characteristics.

Raman hyperspectral imaging is a technique of choice for assessing the distributional homogeneity of compounds of interest. Indeed, the combination of both spectroscopic and spatial information provides a detailed knowledge of chemical composition and component distribution. When dealing with hyperspectral imaging, multivariate data analysis is necessary to extract the concentration map of the compound of interest that will be used to assess sample homogeneity.

Actually, most authors assess homogeneity using parameters of the histogram of intensities (e.g. mean, skewness and kurtosis). However, this approach does not take into account spatial information and loses the main advantage of imaging. Recently, Rosas et al. proposed a homogeneity index based on the Poole index. However, it necessitates cutting the maps in non-overlapping macropixels and is therefore quickly limited with small maps.

To overcome this limitation, we propose a new criterion that combines Continuous Level Moving Blocks and homogeneity curves with a randomization step to assess the distributional homogeneity. This distributional homogeneity index (DHI) enables analysis of hyperspectral maps without apriori knowledge. It has been applied on five pharmaceutical formulations with different blending conditions. The uniformity content values of the API (present at a concentration of 7% w/w) measured by HPLC ranged from RSD: 0.46% to 11.04%.
Ten tablets per formulation have been mapped over a region of interest of 4 mm². After extracting pure spectra by MCR-ALS, the concentration maps of the API were computed using classical least squares analysis. DHI have been computed with a hundred simulations for the randomization step for each concentration map. Afterwards, a mean DHI and standard deviation values were computed per formulation.

A linear relationship has been observed between the RSD values and the mean DHI. These results enabled us to select the formulation with the best homogeneity. Further experiments are in progress to check whether hyperspectral imaging combined with DHI could be used in routine to assess blending homogeneity of well-known formulations.