Novel Fabrication and Optimization Strategy of Microfluidic Drug Delivery Device via Direct Laser Writing and Machine Learning

Abstract

Microneedle-based microinjection systems are essential tools in biomedical research andclinical applications, including gene therapy, stem cell transplantation, and drug delivery. How- ever, conventional microneedles face critical limitations such as tip clogging, restricted diffusion, and tissue deformation during penetration. This dissertation proposes a novel fabrication and optimization strategy for microneedle arrays (MNAs) that integrates high-resolution two-photon direct laser writing (DLW) with machine learning (ML)-driven design frameworks. The research is structured around four distinct but conceptually aligned aims: (i) a foundational ML-driven framework is established to compensate for dimensional inaccuracies in customized macro-scale medical devices. This demonstrates a robust methodology for ”first-time-right” additive man- ufacturing, establishing a computational foundation with high potential for future micro-scale translation. Transitioning directly to the complexities of the micro-domain, the research ad- dresses MNA performance by: (ii) engineering anti-clogging microneedle architectures using DLW; (iii) designing hybrid DLP-DLW printed MNAs to enhance delivery uniformity; and (iv) minimizing tissue deformation through ML-guided MNA optimization. Experimental results demonstrate the high efficacy of this multi-pronged approach. Specif-ically, the ML-driven inverse geometry compensation framework successfully achieves first-time- right manufacturing by significantly reducing dimensional deviations in patient-specific surgical guides. At the micro-scale, the novel 3D-architected microneedles successfully mitigate tip- clogging during in vivo zebrafish embryo microinjections. Furthermore, the hybrid DLP-DLW fabricated MNAs prove mechanically robust during ex vivo mouse brain penetrations, successfully delivering fluidic payloads with enhanced spatial uniformity compared to single-needle systems. Finally, the integration of finite element analysis (FEA) and Bayesian Optimization establish a powerful computational pipeline to actively minimize the force required to puncture brain tissue. Collectively, this platform enables the rapid, customizable fabrication of MNAs tailored to spe- cific anatomical and clinical constraints, offering a transformative pathway for precision-targeted therapeutic delivery.