Quantitative Ultrasound Stochastic Tomography - Revolutionizing breast cancer diagnosis and screening with supercomputing-based radiation-free imaging

通过在地球物理成像领域开发的算法，可以深度增强超声成像。这种基于伴随状态建模和迭代优化的算法提供了非常高分辨率的人体组织的定量图像。目前，这样的图像只能通过高性能计算和使用特定的超声数据采集设备来获得。硬件和软件结合在一起，对软组织成像具有巨大的影响潜力，例如在乳腺癌成像中。然而，正如医学成像中的惯例一样，所获得的图像仅提供每个像素处的组织的平均值或最有可能的值，这是一种非常昂贵的不确定性量化过程，通常被认为是不可行的。基于伴随的超声成像的一项革命性发展使我们有可能以单一的平均值图像为代价获得不确定的图像。这样的发展将是在诊断的置信度估计方面产生变革性影响的基础。我们的目标是通过一种安全(无辐射)、准确(定量)和可靠(不确定性感知)的新型乳腺成像方式来颠覆乳腺癌筛查模式。在QUSTom中，我们将研究基于伴随的不确定性成像背后的基础科学，并确定其在乳腺癌诊断中的潜在适用性。这项技术作为诊断工具的可行性取决于1)采用数据采集硬件以获得最佳分辨率，2)在高性能计算机中实现算法以获得较短的解决方案，以及3)由专业放射科医生与乳房成像最先进的技术进行比较的可行性分析。这一建议涵盖了这三个方面，并开启了将类似原理应用于其他成像领域的可能性，无论是在医学上还是在其他领域。

Ultrasound imaging can be deeply enhanced by means of algorithms developed in the field of geophysical imaging. Such algorithms, based upon adjoint-state modelling and iterative optimization, provide quantitative images of human tissue with very high resolution. At present time, such images can only be attained by means of high-performance computing and using specific ultrasound data acquisition devices. When combined, hardware and software have a huge impact potential for soft-tissue imaging, such as in breast cancer imaging. Nevertheless, and as is customary in medical imaging, the obtained images only provide with the mean, or most likely, values of tissue at each pixel, being uncertainty quantification an extremely expensive process, typically deemed as unfeasible for practical purposes. A revolutionary development in adjoint-based ultrasound imaging allows us to potentially obtain images of uncertainties at the cost of a single, mean-value, image. Such development will be the basis of transformative implications in terms of confidence-estimates for diagnosis. We aim at disrupting the breast cancer screening paradigm by means of a safe (radiation-free), accurate (quantitative) and reliable (uncertainty-aware) novel breast imaging modality. Within QUSTom we will investigate the fundamental science behind adjoint-based uncertainty imaging and establish its potential suitability for breast cancer diagnosis. The feasibility of the technology as a diagnosis tool relies on 1) adapting the data acquisition hardware for optimal resolution, 2) implementing the algorithms in high performance computers in order to obtain a short time-to-solution and 3) feasibility analysis by expert radiologists in comparison with the state-of-the-art in breast imaging. This proposal covers the three aspects and opens the possibility of applying similar principles in other imaging fields, both in medicine and elsewhere.