Generative Model Enhanced X-ray Computed Tomography Imaging for Pharmaceutical Powder Microstructure Reconstruction...

Area of Science:

• Pharmaceutical Manufacturing
• Materials Science
• Computational Modeling

Background:

• Microstructural characteristics and particle interactions significantly impact pharmaceutical manufacturing performance, affecting flowability and dissolution.
• Traditional methods for analyzing microstructures are time-consuming and prone to information loss.
• Variations in microstructure, even with identical compositions, can lead to different process outcomes and product quality.

Purpose of the Study:

• To introduce an innovative production framework for pharmaceutical manufacturing.
• To develop a generative model (Slice to Volume GAN) for creating synthetic microstructures.
• To create a learning-based flowability prediction model using generated microstructure data.

Main Methods:

• Utilized a Slice to Volume GAN (S2V-GAN) with a U-net generator and 2D Slice discriminator.
• Employed a Wasserstein strategy for generating synthetic microstructures from limited X-ray Computed Tomography (XCT) data.
• Developed a flowability prediction model trained on a particle morphology-flow pairing database.

Main Results:

• The S2V-GAN generated high-fidelity 3D microstructures, reducing processing time and providing rich digital data.
• The generative model accurately reconstructed microstructures, showing high similarity in morphological parameter distribution.
• The integrated framework achieved higher accuracy in predicting flowability compared to traditional methods.

Conclusions:

• The developed production framework successfully reconstructs and augments 3D microstructures.
• This approach enables efficient structural analysis and process optimization in pharmaceutical manufacturing.
• The study demonstrates a complete workflow from microstructure generation to flowability prediction.