HOMOGENEITY INDEX FOR THE ASSESSMENT OF HOMOGENEITY IN IMAGING ANALYSIS

Powder mixing is a common part of a manufacturing process in the pharmaceutical, cosmetic and food industry. Mixing is a critical step to maintain a correct distribution of the compound in the formulation. Homogeneity occurs when all the particles are distributed for the entire surface, in a stochastic way. NIR hyperspectral images (NIR-HI) and chemometrics methods are useful to verify the compounds distribution in the surface of a sample. The objective of this work is to comprehensively study the use of continuous-level moving block (CLMB) theory, based on the global standard deviation for each window size (homogeneity curve), and the theories related to homogeneity distribution and scrutiny scale (sampling size) in order to propose a standard homogeneity index. Binary samples were simulated mimicking different condition of mixing: (1) perfect random mixing with adequate particle size; (2) inhomogeneity; and (3) perfect random mixing with different particle sizes. In the first step, with a particle size of 1 squared pixel, different concentrations of active were evaluated (1 – 95%), and an equation were achieved for predict the best homogeneity condition. The equation has two main parameter P1 and P2, and a constant q1. In the inhomogeneity case (II), different concentration (1 – 50%), were verified and the results show a high deviation from the ideal homogeneity curve, as expected. In the third simulation study (III), distinct clump sizes (1, 2, 3, 4, 5, 7 and 10 squared pixel) were evaluated with different concentration (1 – 50%) to verify the scrutiny scale, using the parameters defined by the equation mentioned. Finally, an equation was proposed to evaluate the homogeneity percentage (%H) in real samples, using the area under the curves for homogeneous and non-homogenous distribution, which are the upper and lower model limits, respectively. NIR-HI were acquired by mixing two compound (starch and sucralose), and MCR-ALS were used to obtain the concentration distribution maps (CDM). H% values were estimated using directly the concentration obtained from the CDM as well by binarizing these concentration values. Binarized samples shows more suitable results for %H. Concerning all results, some considerations can be made: (I) the random distribution are the most homogeneous; (II) in real samples, the scrutiny scale cannot be higher than 1 squared pixel, to avoid particles clumps; (III) binarized samples were more suitable to estimate H%; (IV) the mathematical model were suitable to predict homogeneous mixing, concerning the results achieved by CLMB analysis, and the parameters had to be evaluated to verify the model fitting. Further studies have to be made to evaluate the estimation of the H% using directly the concentrations distribution maps.