AI-accelerated optical simulation for fast timing nuclear imaging

用于快速核成像的人工智能加速光学模拟

• 批准号: 10744626
• 负责人: Emilie Roncali
• 金额: $ 60.5万
• 依托单位: UNIVERSITY OF CALIFORNIA AT DAVIS
• 依托单位国家: 美国
• 财政年份: 2023
• 资助国家: 美国
• 起止时间: 2023-08-15 至 2027-07-31
• 项目状态: 未结题
• 来源: https://reporter.nih.gov/project-details/10744626
• 关键词: Acceleration; Address; Affect; Algorithms; Automobile Driving; Clinical; Collaborations; Collection; Communities; Consumption; Custom; Data; Development; Dimensions; Discipline of Nuclear Medicine; Event; Explosion; Family; Geometry; Grant; Image; Image Analysis; Imaging technology; Individual; International; Ionizing radiation; Light; Methodology; Methods; Modeling; Monte Carlo Method; Nanostructures; Nature; Noise; Optics; Performance; Photons; Physics; Positron-Emission Tomography; Process; Production; Pythons; Radio; Research; Research Personnel; Resolution; Signal Transduction; Speed; Structure; System; Technology; Time; Training; United States National Institutes of Health; Work; commercialization; deep learning; design; detector; generative adversarial network; high energy physics; improved; innovation; learning materials; nanophotonic; next generation; novel; nuclear imaging; open data; particle; photonics; radiation detector; reconstruction; simulation; simulation software; single photon emission computed tomography; technology development; theranostics; tool

Project Summary/ Abstract Improving quantification at high spatial resolution is driving technology developments in nuclear imaging. Positron emission tomography (PET) scanners and single photon emission computed tomography (SPECT) use radiation detectors, which performance can be improved when the image formation process is understood. Optimizing optical mechanisms such as scintillation or prompt photon emission at the core of these detectors is essential to advance the technology and is the focus of this proposal. Due to the complexity of these phenomena and the difficulty to disentangle their components experimentally, research on radiation detector optics relies on simulations integrating high and low energy physics. No simulators currently offer the speed and fidelity necessary to understand image formation from the detector to the system. We propose to develop a radically different AI-based high-fidelity optical modeling framework, allowing multidimensional optical information to be rapidly generated, collected, and processed at the system level. By replacing individual photon tracking with a deep-learning approach, we expect to accelerate simulations by several orders of magnitude in systems involving extensive optical photon tracking, such as large detectors or fast timing detectors. We organize this R01 proposal in three specific aims focusing on implementing this framework in the Geant4/GATE simulators and applying it to time-of-flight (TOF) PET. GATE is a free opensource platform at the forefront of nuclear medicine simulation. We have a track record of developing optical modeling strategies and created the LUT Davis model. This grant will design and implement the optiGAN, a custom generative adversarial network (GAN) that will be trained with high-fidelity simulations based on the LUT Davis model. New light transport features and crystal-photodetector interface models mixing particle and wave optics will be developed and integrated into the optiGAN (Aims 1 and 2). We have extensively studied and developed Cerenkov-based radiation detectors, one of the prompt photon emission mechanisms most pursued to achieve timing resolution below 50 ps and unlock reconstruction-free PET. To develop prompt photon-based PET systems several questions must be solved: how to improve the production and transport of these prompt photons with new materials, how to improve their collection, and how to harness the prompt photon information for fast coincidence timing. These questions motivate the development of innovative detector optics and algorithms for TOF PET, which we will investigate with the optiGAN together with experimental and theoretical work (Aims 2 and 3). The objective of this grant is to enable a leap in detector technology through unprecedented simulation capabilities and new strategies to leverage fast detectors in nuclear imaging scanners. Developing detector technology now that enables the next generation of scanners to respond to clinical and research needs of nuclear medicine is essential, as the integration of these advances requires years before commercialization.

项目总结/摘要 在高空间分辨率下改进量化正在推动核成像技术的发展。 正电子发射断层扫描（PET）和单光子发射计算机断层扫描（SPECT） 使用辐射检测器，当图像形成过程被 明白优化光学机制，如闪烁或快速光子发射的核心， 这些探测器对推进技术至关重要，也是本提案的重点。由于复杂性 由于这些现象以及难以通过实验来解开它们的组成部分， 探测器光学依赖于集成高能和低能物理的模拟。目前没有模拟器提供 理解从探测器到系统的图像形成所需的速度和保真度。 我们建议开发一个完全不同的基于AI的高保真光学建模框架， 允许多维光学信息在系统中快速地产生、收集和处理 水平通过用深度学习方法取代单个光子跟踪，我们希望加速 在涉及广泛的光学光子跟踪的系统中， 大型检测器或快速定时检测器。我们将R 01提案分为三个具体目标，重点是 在Geant 4/GATE模拟器中实现该框架，并将其应用于飞行时间（TOF） 彼前GATE是一个处于核医学模拟前沿的免费开源平台。我们有一个轨道 开发光学建模策略的记录，并创建了LUT戴维斯模型。该基金将设计和 实现optiGAN，这是一个定制的生成对抗网络（GAN），将以高保真度进行训练 基于LUT Davis模型的仿真。新的光传输特性和晶体-光电探测器接口 将开发混合粒子和波动光学的模型，并将其纳入optiGAN（目标1和2）。 我们已经广泛研究和开发了基于切伦科夫的辐射探测器， 光子发射机制最追求实现低于50 ps的定时分辨率和解锁 无重建PET。为了开发快速的基于光子的PET系统，必须解决以下几个问题： 如何用新材料改善这些瞬发光子的产生和传输，如何改善 它们的收集，以及如何利用提示光子信息进行快速符合计时。这些 这些问题激发了TOF PET创新探测器光学器件和算法的开发，我们将 与optiGAN一起研究实验和理论工作（目标2和3）。 这项资助的目的是通过前所未有的模拟实现探测器技术的飞跃 在核成像扫描仪中利用快速探测器的能力和新战略。显影检测器 现在的技术，使下一代扫描仪，以响应临床和研究的需要， 核医学是必不可少的，因为这些先进技术的整合需要数年时间才能实现商业化。