Spatiotemporal Prediction of Targeted Nanomedicine Kinetics and Drug Release Using Inverse Physics-Informed Deep Learning and Quantitative MRI
Abstract
Purpose
The clinical translation of liposome-based chemotherapy is limited by heterogeneous drug accumulation and the barriers of the tumor microenvironment (TME). While doxorubicin loaded into nanocarriers exploits the EPR effect, its therapeutic efficacy is restricted by a steep 50 µm perivascular penetration gradient, which leaves distal quiescent cells in the tumor core under-exposed. This project aims to develop a computational framework that integrates Inverse Physics-Informed Neural Networks (I-PINNs) with multi-modal quantitative Magnetic Resonance Imaging to predict the micro-scale distribution and release kinetics of targeted nanomedicines.
Methods
The I-PINN framework embeds known physical laws represented by coupled partial differential equations for encapsulated and free drug pools. This allows the model to predict unknown parameters, such as interstitial fluid pressure and drug release rates, directly from non-invasive imaging data. We will utilize DCE-MRI to map vascular permeability and CEST-MRI to derive spatially resolved extracellular pH maps, which serve as mechanistic triggers for pH-sensitive liposome destabilization. The framework will be validated through in vivo experiments comparing free DOX, non-triggered liposomes, and thermosensitive liposomes under thermal activation, while investigating the potential for radiation therapy to modulate TME fluid dynamics and improve nanoparticle penetration. Projected
Results
It is hypothesized the I-PINN framework will accurately solve the inverse transport problem by precisely identifying constitutive parameters from minimal pre- and post-administration imaging timepoints. The model is expected to quantify the spatiotemporal drug concentration with high fidelity, demonstrating that triggered release mechanisms can bridge the 50 µm perivascular penetration limit to achieve lethal drug exposure in quiescent tumor regions. We anticipate that radiation-induced modulation of TME fluid dynamics will significantly enhance the delivery efficiency of the liposome carriers.
Conclusion
This work represents a significant step toward personalized nanomedicine by providing an imaging-guided computational framework capable of ensuring lethal drug exposure to both proliferating and quiescent tumor cell populations.