In vivo computation with sensor fusion and search acceleration for smart tumor homing

Motivated by the advancements on bioresorbable nanoswimmers, this paper considers the advantages of direct targeting over systemic targeting for smart tumor homing under the general framework of computational nanobiosensing. Nanoswimmers assembled by magnetic nanoparticles can be used as contrast agents to estimate the locations of tumors inside the human body.

Closely observing the response of nanoswimmers (which act as in vivo biosensors) to the tumor-triggered biological gradients and then guiding them through external manipulation, can result in a higher accumulation at the diseased location. Sensor informatics along with data fusion can play a crucial role in such a knowledge-aided targeting process. Specifically, built upon our previous work on direct targeting inspired by the gradient descent optimization, this work is focused on resolving the real-life constraints of in vivo natural computation such as uniformity of the magnetic field and finite life span of the nanoswimmers. To overcome these challenges, we propose a multi-estimate-fusion strategy to obtain a common steering direction for the swarm of nanoswimmers.

We show through computational experiments (1) that the mean of individual gradient estimations provides the best choice for symmetrical conditions (tumor location in line with the direction of blood flow) while leader-based swarm steering gives the best results for non-symmetrical search space, and (2) that the iterative memory-driven gradient descent optimization detects the target faster compared to the classical memory-less gradient descent and knowledge-less systemic targeting.

Our proposed strategies demonstrate that a clear demarcation between malignant tumors and healthy tissues can be visualized before nanoswimmers are consumed in human vasculature. We believe that our work will help in overcoming the challenges posed by natural in vivo computation for tumor diagnosis at its early stage.