(94f) Real-Time Event Tracking in Crystallization with AI-Enabled in-Situ Image Analysis and Morphological Metrics

Crystallization processes are governed by a complex interplay of mechanisms such as nucleation, growth, secondary nucleation, agglomeration, breakage, polymorphism, and oiling out. Accurately detecting and interpreting these changes in real time is essential for robust process development, optimization, and scale-up. In this study, we present a novel AI-enabled toolkit that integrates a convolutional neural network (CNN)-based segmentation engine with in-situ imaging to enable automated, real-time tracking of crystallization events. The toolkit extracts and monitor time-resolved morphological descriptors—including aspect ratio, circularity, roughness factor, particle length, and circular equivalent perimeter etc—from segmented particles in live process images. These quantitative shape- and size-based metrics form the foundation for a data-driven event tracking approach, allowing precise characterization of process changes such as nucleation onset, habit transformation, and agglomeration.

The application was deployed in a combined cooling and antisolvent crystallization system, where it successfully captured key process transitions: primary nucleation during cooling, desupersaturation and secondary nucleation during the hold phase, and agglomeration and crystal habit changes during antisolvent addition. In two case studies, insights derived from the platform enabled significant process improvements—including reduction of the antisolvent addition phase and a shift in crystal habit from needle-like to plate-like forms, resulting in markedly enhanced filtration performance. The AI-enabled analysis offers consistent, high-resolution morphological tracking at a speed that supports near real-time decision-making. This work demonstrates how the integration of deep learning with in-situ image analysis—driven by shape- and size-based metrics—can transform crystallization process development from qualitative observation to real-time, quantitative control. The platform enhances mechanistic understanding, accelerates process optimization, and strengthens the foundation for robust tech transfer and scale-up in pharmaceutical crystallization.