Combining Machine Learning and Backgrounded Membrane Imaging: A Case Study in Comparing and Classifying Different...

Area of Science:

• Biopharmaceutical analysis
• Particle characterization
• Image analysis

Background:

• Backgrounded membrane imaging (BMI) offers advantages for studying subvisible particles, including low sample volume requirements and avoidance of common imaging issues.
• The morphological information within brightfield BMI images has been largely underutilized for particle analysis.
• Characterizing subvisible particles is crucial throughout drug product development.

Purpose of the Study:

• To investigate the use of convolutional neural networks (CNNs) for quantitative and qualitative analysis of subvisible particles using BMI.
• To demonstrate the capability of CNNs in extracting morphological information from label-free brightfield BMI images.
• To assess the utility of CNN-based methods for distinguishing various particle types in biopharmaceuticals and model systems.

Main Methods:

• Analysis of particle images from biopharmaceutical products and laboratory-prepared samples using two CNN approaches: supervised classifiers and a fingerprinting analysis method.
• Application of CNNs to extract morphological features from brightfield BMI images.
• Utilizing a fingerprinting method for comparative analysis of particle morphology.

Main Results:

• CNN-based methods effectively extract morphological information from label-free brightfield BMI particle images.
• The CNN approaches successfully distinguish between particles composed of different proteins, fatty acids, and protein surrogates (NIST ETFE) based solely on BMI images.
• The fingerprinting method proved useful for comparing morphological similarities and differences across various particle types and sample origins.

Conclusions:

• CNNs are powerful tools for leveraging morphological data from BMI, enabling detailed particle characterization.
• This approach enhances the ability to identify and differentiate subvisible particles in biopharmaceutical development.
• The study highlights the potential of combining BMI with CNNs for advanced particle analysis in drug product quality control.