Machine Learning-Enabled Image Comparability Assessment for Flow Imaging Microscopy Across Platforms

Area of Science:

• Biopharmaceutical analysis
• Analytical chemistry
• Microscopy techniques

Background:

• Biopharmaceutical development necessitates stringent quality control, with subvisible particles being a critical attribute.
• Current pharmacopoeial methods (Light Obscuration, Microscopic Particle Count) have limitations for advanced analytical needs.
• Flow Imaging Microscopy (FIM) presents a potential alternative, but standardization is ongoing.

Purpose of the Study:

• To evaluate the consistency and transferability of particle counting across different Flow Imaging Microscopy (FIM) platforms.
• To assess the feasibility of standardized subvisible particle analysis using FIM technology.
• To explore image classification consistency for FIM data across manufacturers.

Main Methods:

• Instrument standardization and consistency evaluation using polystyrene microsphere standards and intravenous immunoglobulin.
• Parallel testing of FIM instruments from different manufacturers.
• Particle image classification using convolutional neural networks and UMAP dimensionality reduction.

Main Results:

• A transfer strategy for consistent particle counting results across FIM platforms was developed.
• Particle image classification demonstrated consistency in categorization but highlighted cross-platform recognition challenges.
• The study confirmed the potential for reliable subvisible particle analysis with FIM.

Conclusions:

• Findings contribute to the standardization of Flow Imaging Microscopy for biopharmaceutical quality control.
• FIM shows promise as a reliable analytical tool for subvisible particle characterization.
• Further development is needed to fully standardize cross-platform image recognition in FIM.