(610g) Imaging Crystallization Using Deep Learning to Quantitatively Track the Polymorphic Transformation of Carbamazepine | AIChE

(610g) Imaging Crystallization Using Deep Learning to Quantitatively Track the Polymorphic Transformation of Carbamazepine


Gao, Z. - Presenter, Tianjin University
Rohani, S., Western University
Wu, Y., Western University
Gong, J., Tianjin University
Bao, Y., Tianjin University
Wang, J., National Engineering Research Center for Industrial Crystallization Technology, School of Chemical Engineering and Technology, Tianjin University
Solvent mediated polymorphic transformation (SMPT) is a crucial phenomenon for a specific polymorph crystals’ production and stability control in pharmaceutical crystallization processes. In this study, a completely new method based on computer vision was developed using deep learning to quantitatively track crystal individuals. For the model compound carbamazepine (CBZ), two polymorphic forms with different morphologies were identified using deep learning image segmentation. The geometry models of two crystal forms were built to calculate the volume of each crystal in the real-time tracking process. The spatial valid vision volume of a high-speed camera probe was calculated and then verified by glass bead. Furthermore, the mass of two polymorphic crystals in suspension were fitted by f1 and f2 through comparing real-time tracking and off-line measurement. At last, the real-time quantitative tracking of polymorphic transformation of CBZ was realized through deep learning-based imaging analysis. The transformation process including dissolution of form II, nucleation and growth of form III, crystal size distribution (CSD) etc. was analyzed and identified by off-line measurement. Our work in this study attempts to bridge the gap between the advanced image tracking technology available today and the specific needs of solution crystallization to track, count and measure the crystal individuals. It is a great opportunity to proceed intelligent manufacturing through visualizing traditional crystallization process with the help of artificial intelligence (AI).


The authors acknowledge the financial support provided by the Natural Science and Engineering Research Council (NSERC) of Canada, National Natural Science Foundation of China (NNSFC 21576187, NNSFC 21776203, NNSFC 21621004, NNSFC 81361140344), and the China Scholarship Council.