FAIR-CT: a practical approach to enable ultra-low dose CT for longitudinal disease and treatment monitoring

项目摘要/摘要 超低剂量CT，定义为整个胸部、腹部或骨盆的亚毫西弗(亚毫西弗)成像，是 对于慢性病和癌症患者的医疗保健来说是至关重要的。不幸的是，光子饥饿 电子噪声使得在这样的剂量水平下进行成像具有挑战性。光子饥饿指的是 传输的光子。当没有光子被传输时，测量基本上是无用的。如果只有几个光子 被传输，测量携带信息，但其解释和价值被混淆 电子噪音。已经提供了用于亚mSv胸部成像的解决方案，结果令人鼓舞，但这些 不是广泛可用的，也不容易在解剖部位、供应商和扫描仪型号之间推广。 我们提出了一种新的、健壮的超低剂量CT解决方案，它将克服这些问题。我们是指我们的 解决方案为FIRE-CT，它代表有限角度集成射线CT。FIRE-CT在以下原则下运行 光子匮乏和电子噪音的混杂效应最好是通过避免它们来处理，这 可以通过在源-探测器旋转期间增加数据积分时间来实现。FIRE-CT数据 强烈背离经典CT数据模型，共享稀疏视图的条纹伪影问题 取样。FIRE-CT数据采集也会影响方位分辨率。我们预计这些问题可能是 使用先进的图像重建技术进行适当处理。一旦可用，FIRE-CT将允许 改善对慢性阻塞性肺病、尿石症和慢性疾病患者的纵向监测 糖尿病，从而减少死亡率和合并症。FIRE-CT还将允许推进癌症治疗 通过允许在不同剂量分数之间调整放射治疗计划来进行治疗 增加CT辐射暴露，并促进早期检测以药物为基础的炎症 治疗。为了使公平-CT取得成果，我们将努力实现两个具体目标：(1)建立一个 全面收集FIRE-CT数据集，实现严格的开发、验证和评估 图像重建算法；(2)先进图像重建算法的发展、验证和评价 算法。FIRE-CT数据集将涉及使用最先进的扫描仪，并包括REAL 从高剂量扫描合成的患者数据用于标准护理。两幅互补图像 将对重建方法进行调查。即基于模型的非线性迭代重建 前向模型和专用压缩感知正则化；基于深度学习的FBP精化 使用具有任务适应的图像质量的目标图像进行重建。图像质量评估将占到 关键生物变量，涉及结构相似性和对比度噪声比等客观指标 用于临床证实的损伤，以及涉及人类读者的基于任务的性能指标。

Project Abstract/Summary Ultra-low dose CT, defined as sub-millisievert (sub-mSv) imaging of the entire chest, abdomen or pelvis, is critically needed for healthcare of patients with chronic diseases and cancer. Unfortunately, photon starvation and electronic noise make imaging at such dose levels challenging. Photon starvation refers to the number of transmitted photons. When no photons are transmitted, the measurement is essentially useless. If few photons are transmitted, the measurement carries information, but its interpretation and value are confounded by electronic noise. Solutions with encouraging results have been offered for sub-mSv chest imaging, but these are not widely available and not easily generalizable across anatomical sites, vendors and scanner models. We propose a novel, robust solution for ultra-low dose CT that will overcome these issues. We refer to our solution as FAIR-CT, which stands for Finite-Angle Integrated-Ray CT. FAIR-CT operates under the principle that photon starvation and the confounding effect of electronic noise are best handled by avoiding them, which is made possible by increasing the data integration time during the source-detector rotation. FAIR-CT data strongly deviate from the classical CT data model and share the streak artifact problem of sparse view sampling. FAIR-CT data acquisition also affects azimuthal resolution. We anticipate that these issues can be suitably handled using advanced image reconstruction techniques. Once available, FAIR-CT will allow improvements in longitudinal monitoring of patients with chronic diseases such as COPD, urolithiasis and diabetes, thereby reducing mortality and co-morbidities. FAIR-CT will also allow advancing cancer therapy treatments by enabling adjustments in radiation therapy plans between dose fractions without increasing CT radiation exposure, and by facilitating early detection of inflammations in drug-based therapies. To bring FAIR-CT towards fruition, we will work on two specific aims: (1) Creation of a comprehensive collection of FAIR-CT data sets enabling rigorous development, validation and evaluation of image reconstruction algorithms; (2) Development, validation and evaluation of advanced image reconstruction algorithms. The FAIR-CT data sets will involve the utilization of state-of-the-art scanners and include real patient data synthesized from high dose scans acquired for standard of care. Two complementary image reconstruction approaches will be investigated. Namely, model-based iterative reconstruction with non-linear forward model and dedicated compressed sensing regularization; and deep learning-based refinement of FBP reconstructions using target images with task-adapted image quality. Image quality evaluation will account for critical biological variables and involve objective metrics such as structure similarity and contrast-to-noise ratio for clinically-proven lesions, as well as task-based performance metrics involving human readers.