Direct Image Feature Extraction and Multivariate Analysis for Crystallization Process Characterization.

Area of Science:

• Chemical Engineering
• Crystallization Process Development
• Image Analysis

Background:

• Small-scale crystallization experiments (1-8 mL) are crucial for early-stage process development, providing data on solubility and kinetics.
• Digital imaging is commonly used for monitoring, offering qualitative insights or object detection for size and shape analysis.

Purpose of the Study:

• To present a novel approach for routine characterization of image data from crystallization experiments using direct image feature extraction.
• To apply extracted image features for accurate clear/cloud point detection and crystal suspension density prediction.

Main Methods:

• Extraction of 80 image features using image statistics, histogram parametrization, and image transformations to assess grayscale characteristics.
• Utilized features for clear/cloud point detection and crystal suspension density prediction via partial least-squares regression (PLSR).

Main Results:

• The image-based method achieved significantly higher accuracy for clear/cloud point detection (MAE 0.42 mg/mL) compared to transmission-based methods (MAE 8.99 mg/mL).
• PLSR models successfully predicted crystal suspension densities up to 40 mg/mL (R² > 0.81, Q² > 0.83).

Conclusions:

• The developed image analysis methodology provides accurate, quantitative measurements for early parameter estimation and process modeling in crystallization.
• This approach has potential for broader application in monitoring key physical parameters across various imaging techniques for particle and crystallization processes.