Geometry driven machine learning guided fabrication of sub-5 µm PLGA microparticles for balloon catheter drug delivery.

Poly (lactic-co-glycolic acid) (PLGA) microparticles are widely used as biodegradable drug-delivery carriers, where particle size critically governs drug release kinetics, degradation behavior, and overall therapeutic performance. Producing sub-5 µm PLGA microparticles in PDMS microfluidic devices remains challenging due to limited shear control, PDMS swelling in organic solvents, and instability at high flow-rate ratios. This study demonstrates stable generation of sub-5 µm PLGA microparticles by integrating a shear-modulating obstruction into a cross-junction microfluidic device to enhance droplet breakup under laminar flow. Computational Fluid Dynamics (CFD) simulations and initial experiments showed that positioning a 150 µm-wide obstruction 5 µm downstream shifted droplet formation into a stable dripping regime, reducing droplet size from 84 ± 5 µm (10 : 1) and 32 ± 3 µm (20 : 1) in the unobstructed device to 8 ± 0.7 µm with controlled pinch-off, followed by a machine-learning-guided optimization model to identify parameter combinations that yielded 3.6-4.5 µm microparticles under optimized conditions. A Polynomial Regression surrogate model in a machine learning-assisted optimization accurately captured nonlinear interactions between flow-rate ratio, obstruction width, and obstruction distance. Coupling the model with randomized grid search (n = 2000) yielded design combinations that experimentally reproduced particle sizes within <3% error for four of five candidate models. Finally, a thermoresponsive biodegradable polymer adhesive enabled temperature-activated attachment of the unloaded PLGA microparticles to a coronary balloon catheter and partial particle transfer during balloon inflation in a simplified silicone arterial model, providing an enabling platform for future catheter-based drug-delivery studies.