(660c) Computer Vision and Image Processing for Particle Size Estimation from Images Collected during Crystallization Design

Particle size distribution (PSD) of small molecule active pharmaceutical ingredients (APIs) is of critical importance and extensively controlled during crystallization process development. The PSDs of the final API material significantly affect the dissolution, flow, and processability of the material in drug product development. PSDs are typically measured using offline instruments and are solvent and time intensive measurements. PSDs of the isolated solids are used to iteratively tune crystallization design considerations such as, seeding point, wet-milling strategy, incorporation of temperature cycle etc. to target a specific range of PSDs. Additionally, very early-stage API programs typically do not have an established experimental protocol for offline measurements. This is due to the lack of information on the refractive index of crystals, propensity of breakage under shear etc. in the nascent stages of development. Accordingly, it is beneficial to leverage a quicker and efficient method of estimating PSDs that does not completely rely on repetitive offline measurements.

In this work, we develop a workflow for estimating particle size distributions from Polarized Light Microscopy (PLM) images that are routinely captured during crystallization experiments. We leverage publicly available open computer vision libraries and build a workflow that can be easily democratized. The method allows statistical analysis of PLM images and maximizes the utilization of information from existing experimental data. We demonstrate that the predictions from the computational workflow are in good agreement with the experimental measurements of PSDs for multiple API molecules. This framework may be advanced to other data formats such as videos captured by PAT tools to provide an in-line estimation of PSDs. We show that the framework can be executed in real time and successfully makes predictions for D50 and D90 of crystals with distinct morphologies.

Disclosure:

Kartik Kamat is an employee of AbbVie. All authors may own AbbVie stock. AbbVie sponsored and funded the study; contributed to the design; participated in the collection, analysis, and interpretation of data, and in writing, reviewing, and approval of the final publication.