Convolutional neural networks for segmentation of FIB-SEM nanotomography data from porous polymer films for controlled drug release | JoVE Visualize

Convolutional neural networks for segmentation of FIB-SEM nanotomography data from porous polymer films for controlled drug release

Fredrik Skärberg¹, Cecilia Fager^2,3, Francisco Mendoza-Lara⁴

¹Bioeconomy and Health, Agriculture and Food, RISE Research Institutes of Sweden, Göteborg, Sweden.

Journal of Microscopy|April 2, 2021

View abstract on PubMed

Summary

Convolutional neural networks accurately segment porous polymer films used in controlled drug release coatings. This advancement improves upon previous methods, enabling better understanding and tailoring of drug delivery systems.

Area of Science:

Materials Science
Pharmaceutical Technology
Computational Imaging

Background:

Phase-separated polymer films, specifically ethyl cellulose and hydroxypropyl cellulose (EC/HPC) blends, are crucial for controlled drug release from oral dosage forms.
Leaching of the hydroxypropyl cellulose (HPC) phase from EC/HPC films creates a porous ethyl cellulose (EC) network essential for drug transport and release rate modulation.

Purpose of the Study:

To implement and evaluate convolutional neural networks (CNNs) for the segmentation of porous EC films derived from FIB-SEM tomography data.
To assess the performance of CNN-based segmentation against manual segmentations and prior machine learning approaches for characterizing porous film structures.

Main Methods:

Acquisition of FIB-SEM tomography data from three porous EC films with varying porosities, relevant to controlled drug release applications.
Development and application of convolutional neural networks for image segmentation to identify and analyze the porous network within the EC films.
Comparison of CNN segmentation results with manual segmentations and a random forest classifier approach.

Main Results:

CNNs demonstrated very good agreement with manual segmentations of the porous EC film structures.
The implemented CNN approach showed improved segmentation performance compared to previous work utilizing random forest classifiers on the same datasets.
The study successfully characterized porous networks in EC films across a range of porosities relevant to pharmaceutical applications.

Conclusions:

Convolutional neural networks offer a robust and accurate method for segmenting FIB-SEM data of porous polymer films used in controlled drug release.
This work advances the characterization of drug delivery systems by providing improved image segmentation tools.
Open-access release of data and software will facilitate further research and development in FIB-SEM segmentation techniques for pharmaceutical applications.