Next-Generation Ultrasound Localization Microscopy

项目摘要/摘要 组织微循环的异常改变往往与组织病理的早期阶段有关。 对这些早期微血管异常进行检测和定性，有助于临床诊断 和治疗监测，以及促进创造新的治疗方法，以对抗疾病的发展。 几十年来，人们一直在寻求开发一种临床成像模式，这种模式可以 非侵入性地直接成像这种组织微血管变异。然而，到目前为止，这种成像 由于成像空间分辨率和深度之间的根本折衷，方法仍然难以捉摸 穿透力。因此，该项目的长期目标是通过开发 新一代超声定位显微镜(ULM)是一种基于超声的成像技术 它可以在临床上直接评估人体体内结构和功能组织微血管的形成 深度。与其他成像方式不同，ULM不受分辨率-穿透度折衷的限制： ULM可以对几厘米深的毛细血管尺度的微血管进行非侵入性成像和定量 测量血流速度(低至1 mm/S)。这样的深度成像穿透力和精致的结合 空间分辨率和测量小血管血流速度的独特功能使ULM成为一种有前途的 这项技术可用于许多临床应用，包括癌症和心血管疾病。然而，目前ULM 由于几个关键技术限制，尚未准备好用于临床：1)ULM数据采集非常缓慢(数十 屏息数秒)；2)ULM后处理在计算上非常昂贵(几个小时 生成单个2D ULM图像)；3)ULM难以扩展到3D成像(这对于 组织微血管的综合评价，例如在癌症中的应用)。这些限制在很大程度上 禁止ULM在临床上有效地用于提供有用的微血管生物标志物。在这份提案中， 我们将集中精力解决这些技术障碍，并将ULM转变为真正有用的临床成像 工具。我们的方法协同结合了深度学习(DL)、并行计算和超高速3D超声 成像可从根本上缩短ULM数据采集时间，大幅加快ULM后处理速度，以及 增强ULM到3D的成像。我们的首要目标是开发和验证基于DL的ULM数据处理算法 这将使实时4D形态ULM和快速3D定量ULM成为可能。我们的方法独特地收集了 用于DL训练的鸡胚微血管模型上的真实标记光学成像数据。我们的第二个目标是 重点研究了利用超快三维平面波成像技术在二维行-列寻址换能器上实现三维超视距。 我们将开发一种基于DL的波束形成技术，以实现高保真3D微泡成像 3D-ULM。我们的最终目标将集中在验证新开发的3D-ULM成像的体内性能 关于小鼠肿瘤模型的技术。我们将与世界知名的深度学习专家合作， 光学成像，以及伊利诺伊大学的比较医学来实现这些目标的提案。

Project Summary/Abstract Abnormal alterations of tissue microcirculation are often associated with early stage of tissue pathology. Detection and characterization of these early microvascular abnormalities can greatly benefit clinical diagnosis and treatment monitoring as well as facilitating the creation of new therapies to counter disease development. For decades, there has been a longstanding quest for the development of a clinical imaging modality that can noninvasively and directly image such tissue microvascular variations. To date, however, such an imaging method remains elusive due to the fundamental compromise between imaging spatial resolution and depth penetration. Therefore, the long-term objective of this project is to fulfill this unmet clinical need by developing the next-generation ultrasound localization microscopy (ULM), which is an ultrasound-based imaging technique that can directly assess structural and functional tissue microvasculature in vivo in humans at a clinically relevant depth. Different from other imaging modalities, ULM is not limited by the resolution-penetration compromise: ULM can noninvasively image capillary-scale microvessels at several centimeters depth and quantitatively measure their blood flow speed (as low as 1 mm/s). Such combination of deep imaging penetration and exquisite spatial resolution and the unique functionality of measuring small vessel blood flow speed make ULM a promising technique for many clinical applications including cancer and cardiovascular diseases. At present, however, ULM is not ready for clinical use due to several key technical limitations: 1) ULM data acquisition is very slow (tens of seconds with breath holding); 2) ULM post-processing is very expensive computationally (several hours to generate a single 2D ULM image); 3) ULM is difficult to be extended to 3D imaging (which is important for comprehensive evaluation of tissue microvasculature such as in cancer applications). These limitations largely forbids ULM from being effectively used in the clinic to provide useful microvascular biomarkers. In this proposal, we will concentrate on addressing these technical barriers and transform ULM to a truly useful clinical imaging tool. Our approach synergistically combines deep learning (DL), parallel computing, and ultrafast 3D ultrasound imaging to fundamentally shorten ULM data acquisition time, substantially accelerate ULM post-processing, and enhance ULM to 3D imaging. Our first aim will develop and validate DL-based ULM data processing algorithms that would enable real-time 4D morphometric ULM and fast 3D quantitative ULM. Our method uniquely collects real labeled optical imaging data on a chicken embryo microvessel model for DL training. Our second aim will focus on realizing 3D-ULM on a 2D row-column-addressing transducer with ultrafast 3D plane wave imaging. We will develop a DL-based beamforming technique to enable high-fidelity 3D microbubble imaging for robust 3D-ULM. Our final aim will focus on validating the in vivo performance of the newly developed 3D-ULM imaging techniques on a mouse tumor model. We will be collaborating with world-renowned experts in deep learning, optical imaging, and comparative medicine at the University of Illinois to accomplish these aims of the proposal.