The consistent production of uniformly sized polymeric microparticles is crucial for their stability and effectiveness in various applications, including drug delivery and biomimetic studies. Traditional bulk methods for producing these particles often suffer from limitations in size uniformity and control. Microfluidics offers a powerful alternative, enabling the generation of monodisperse microparticles with precisely tunable properties and high encapsulation efficiency. However, the optimization of microfluidic systems can be time-consuming and laborious, requiring extensive experimental trials to determine the optimal parameters for achieving desired particle sizes. Predicting the size of PLGA droplets and the resulting particles after solvent evaporation based solely on fluid properties is challenging due to the difficulty of accurately determining parameters such as viscosity and surface tension. This research addresses this challenge by leveraging the power of artificial intelligence (AI), specifically machine learning techniques, to create predictive models for the size of PLGA microparticles produced via microfluidics. The use of AI in various fields, such as pharmaceutics and engineering, has demonstrated its potential to accelerate research and development. Artificial neural networks (ANNs), inspired by biological neurons, are particularly suitable for modeling complex non-linear relationships, making them an ideal tool for predicting particle size in microfluidic systems. This study aimed to develop in silico ANN models capable of predicting the size of PLGA microparticles generated by different microfluidic systems, both individually and in combination, significantly improving the efficiency and speed of microparticle production for various applications.

Numerous studies have explored the application of machine learning in various fields, highlighting its potential for predictive modeling and optimization. In the pharmaceutical industry, AI has been applied to predict drug solubility and other physicochemical properties, streamlining drug development. In the field of microfluidics, machine learning models have been developed to predict droplet size and stability, helping optimize microfluidic device designs and operating conditions. However, these previous studies often focused on a single microfluidic system. The application of AI to predict parameters in microfluidic systems, especially concerning nanoprecipitation of drug particles, droplet stability and fluid flow characteristics has shown promise. This study expands upon this existing work by developing ANN models capable of predicting particle sizes across multiple microfluidic systems.

Three different microfluidic systems (MFS A, MFS B, and MFS C) were used to generate PLGA microparticles. MFS A employed a single-junction 3D flow focusing device to produce single emulsions of PLGA droplets. MFS B utilized a 3D flow focusing chip with seven parallel junctions to increase throughput. MFS C generated multiple emulsions using two sequentially coupled single-junction chips, resulting in core-shell PLGA particles. PLGA microparticles were fabricated by dissolving PLGA in dichloromethane (DCM), introducing it into the microfluidic device along with an aqueous continuous phase, and allowing DCM to evaporate, solidifying the particles. The size and uniformity of the droplets and resulting particles were monitored using high-speed microscopy. The experimental parameters, including PLGA concentration, flow rates of the aqueous and DCM phases, and chip type, were used as input variables for the ANN models. Five different ANN models were developed: ANN-A (MFS A), ANN-B (MFS B), ANN-C (MFS C), ANN-AB (MFS A and MFS B combined), and ANN-ABC (MFS A, MFS B, and MFS C combined). The Statistica v13.3 software was used to implement multilayer perceptron (MLP) ANNs. Various activation functions and numbers of hidden layer nodes were explored to optimize the ANN architecture. The datasets were divided into training, testing, and validation sets to ensure the robustness of the models. The performance of the ANN models was evaluated using the correlation coefficient (r²) and error values.

All five ANN models demonstrated high accuracy in predicting PLGA microparticle size. The individual models (ANN-A, ANN-B, ANN-C) showed excellent performance with r² values above 0.98 for training, testing, and validation datasets. The combined models (ANN-AB and ANN-ABC) also yielded high accuracy, indicating the ANN's ability to generalize across different microfluidic systems. The ANN-ABC model, trained on data from all three systems, achieved r² values of 0.978, 0.975, and 0.971 for the training, testing, and validation datasets, respectively. Sensitivity analysis revealed that the most important factors influencing particle size were aqueous phase flow rate, PLGA concentration, and PLGA flow rate, with chip type also playing a significant role in the combined models. The experimental results showed that increasing PLGA concentration or decreasing aqueous phase flow rate while increasing PLGA disperse phase flow rate led to larger droplets and particles. The sizes of the droplets and particles were tunable by changing the flow rates of the two phases. Microfluidic system A produced droplets with sizes ranging from 35 to 101 µm (1% PLGA) and particles with sizes from 12 to 23 µm after DCM evaporation. Microfluidic system B produced droplets with sizes ranging from 26 to 102.5 µm and particles from 7 to 40 µm. Microfluidic system C produced inner water droplets (3–14 µm), outer PLGA droplets (57–72 µm), and final particles (30–45 µm).

This study successfully demonstrates the potential of AI-powered predictive modeling for optimizing microfluidic particle fabrication. The high accuracy of the developed ANN models, particularly the ANN-ABC model, which integrates data from three different microfluidic systems, showcases the ability of this approach to generalize across various experimental conditions. The insights gleaned from the sensitivity analysis highlight the relative importance of different input parameters in determining particle size, providing valuable guidance for process optimization and control. The development of a single ANN model (ANN-ABC) capable of predicting particle size across multiple microfluidic platforms simplifies the design and optimization process, potentially reducing development time and cost. The successful prediction of particle size using machine learning simplifies the optimization process, significantly reducing the time and resources required to achieve desired particle characteristics. This has implications for the broader field of microparticle production, streamlining the design and optimization of microfluidic devices for various applications.

This work presents a novel application of AI for rapid and efficient fabrication of size-tunable PLGA microparticles using various microfluidic systems. The development of the ANN-ABC model, capable of accurate predictions across three different microfluidic platforms, represents a significant advancement in the field. This AI-driven approach promises to accelerate the development of new microparticle-based technologies for biomedical and pharmaceutical applications. Future research could explore the applicability of this approach to other polymeric materials and microfluidic designs, further expanding its potential impact on various industries.

While the ANN models demonstrated high accuracy, the relatively small size of the datasets used to train some of the models (e.g., ANN-C) might lead to overfitting. The generalizability of the models beyond the specific experimental conditions and materials used in this study needs further investigation. Additional research is needed to explore the scalability of the approach to larger-scale production and different microfluidic geometries.