Next-generation Monte Carlo eXtreme Light Transport Simulation Platform

点击翻译按钮获取中文摘要

Project Summary/Abstract Abstract: The rapid evolution of the field of biophotonics has produced numerous emerging techniques for combatting diseases and addressing urgent human health challenges, offering safe, non-invasive, and portable light-based diagnostic and therapeutic methods, and attracting exponentially growing attention over the past decade. Rigorous, fast, versatile and publicly available computational tools have played pivotal roles in the success of these novel approaches, leading to breakthroughs in new instrumentation designs and extensive explorations of complex biological systems such as human brains. The Monte Carlo eXtreme (MCX, http://mcx.space) light transport simulation platform developed by our team has become one of the most widely disseminated biophotonics modeling platforms, known for its high accuracy, high speed and versatility, as attested to by its over 27,000 downloads and nearly 1,000 citations from a large (2,400+ registered users) world-wide user community. Over the past years, we have also been pushing the boundaries in cutting-edge Monte Carlo (MC) photon simulation algorithms by exploring modern GPU architectures, advanced anatomical modeling methods and systematic software optimizations. In this proposed project, we will build upon the strong momentum created in the initial funding period, and strive to further advance the state-of-the-art of GPU-accelerated MC light transport modeling with strong support from the world’s leading GPU manufacturers and experts, further expanding our platform to address a number of emerging challenges in biomedical optics applications. Specifically, we will further explore emerging GPU architecture and resources, such as ray- tracing cores, half- and mixed-precision hardware, and portable programming models, to further accelerate the MC modeling speed. We will also develop hybrid shape/mesh-based MC algorithms to dramatically advance the capability in simulating extremely complex yet realistic anatomical structures, such as porous tissues in the lung, dense vessel networks in the brain, and multi-scaled tissue domains. In parallel, we aim to make a break- through in applying deep-learning-based image denoising techniques to equivalently accelerate MC simulations by 2 to 3 orders of magnitudes, as suggested in our preliminary studies. In the continuation of this project, we strive to create a dynamic and community-engaging simulation environment by extending our software to allow users to create, share, browse, and reuse pre-configured simulations, avoiding redundant works in re-creating complex simulations and facilitating reproducible research. In addition, we will expand our well-received user training programs and widely disseminate our open-source tools via major Linux distributions and container images. At the end of this continued funding period, we will provide the community with a significantly accelerated, widely-available and well-supported biophotonics modeling platform that can handle multi-scaled tissue optical modeling ranging from microscopic to macroscopic domains.