Optimization of diagnostic accuracy, radiation dose, and patient throughput for cardiac SPECT via advanced and clinically practical cardiac-respiratory motion correction and deep learning

单光子发射型计算机断层扫描（SPECT）心肌灌注成像（MPI）被广泛使用 检测和评估冠状动脉疾病。该项目的目标是减少辐射剂量和/或 SPECT MPI的扫描时间增加了16倍，同时保持或提高了诊断准确性。这 将使得能够执行SPECT MPI，例如，辐射剂量降低4倍，扫描时间缩短4倍（约2.5 分钟）比典型协议。SPECT MPI中的辐射剂量已被认为是重要的 问题，约占医疗成像中患者所有辐射暴露的25%。剂量减低 特别是解决肥胖患者（接受较高剂量）和年轻患者患病率增加的问题。 心脏病患者（由于预期寿命较长，辐射风险较高）。缩短扫描时间 将改善老年人和体弱心脏病患者的舒适度，同时减轻身体运动图像， 人工制品，并通过增加临床吞吐量来降低医疗保健成本。我们将减少剂量和扫描时间 通过创新的图像重建方法，涉及很少或没有成本，不需要额外的 患者设置步骤。我们将采用新的呼吸和心脏运动补偿，以减少图像 人工制品，以及新的深度学习技术，这将用于两个验证信号估计 和高性能降噪。我们将有条不紊地优化这些技术，然后验证我们的 多中心临床阅片师研究中的算法。 SA 1：为低计数研究开发临床实用的呼吸运动替代物。T1：完美的数据- 驱动的呼吸替代估计; T2：在减少的计数下优化数据驱动的替代估计; T3： 开发和临床验证用于呼吸和身体运动替代估计的深度传感相机; T4：将数据驱动的替代估计推广到没有CT的SPECT系统。 SA 2：开发深度学习重建方法并优化诊断准确性和剂量/扫描 时间T1：3D灌注图像的重建后DL去噪算法，用于减少计数和标准- 计数研究; T2：4D心脏门控研究的DL去噪算法; T3：嵌入式4D重建 DL去噪、心脏运动估计和校正;以及T4：使用RMC和 和CMC，其中使用在SA 1中导出的呼吸替代信号对投影数据进行分箱。 SA 3：进行多中心临床阅片人研究（6名临床医生，3家机构），以验证新的 算法，并基于诊断性能与当前临床可用的方法进行比较， 评价灌注和室壁运动缺陷的可重复性。T1：与基线临床 重建，评价以下方面的附加受益：a）包括衰减和散射校正，以及B）另外 包括RMC; T2：验证DL是否可以改善灌注和功能（室壁运动）任务表现 全计数水平;和T3：在减少计数的情况下提高任务性能的T3-DL。

Single-photon emission computed tomography (SPECT) myocardial perfusion imaging (MPI) is widely used to detect and evaluate coronary artery disease. The goal of this project is to reduce the radiation dose and/or scan time of SPECT MPI by a combined factor of 16x, while maintaining or increasing diagnostic accuracy. This would enable SPECT MPI to be performed, e.g., with 4x reduced radiation dose and 4x shorter scan time (~2.5 minutes) than typical protocols. Radiation dose in SPECT MPI has been recognized as an important issue, accounting for ~25% of all radiation exposure to patients in medical imaging. Dose reduction particularly addresses the increased prevalence of obese patients (who receive higher dose) and younger cardiac patients (whose radiation risk is higher due to longer life expectancy). Reduction in scan time would improve comfort for elderly and infirm cardiac patients, while mitigating body-motion image artifacts and reducing healthcare costs by increasing clinical throughput. We will reduce dose and scan time through innovative image reconstruction methods that involve little or no cost and require no additional patient setup steps. We will employ new respiratory and cardiac motion compensation to reduce image artifacts, as well as new deep learning techniques, which will be used for both respiratory-signal estimation and high-performance denoising. We will methodically optimize these techniques and then validate our algorithms in multicenter clinical reader studies. SA1: Develop clinically practical respiratory motion surrogates for low-count studies. T1: Perfect data- driven respiratory surrogate estimation; T2: Optimize data-driven surrogate estimation at reduced counts; T3: Develop and clinically validate depth-sensing cameras for respiratory and body-motion surrogate estimation; T4: Generalization of data-driven surrogate estimation to SPECT systems not having a CT. SA2: Develop deep-learning reconstruction methods and optimize for diagnostic accuracy and dose/scan time. T1: Post-reconstruction DL denoising algorithms for 3D perfusion images for reduced-count and standard- count studies; T2: DL denoising algorithms for 4D cardiac-gated studies; T3: 4D reconstruction with embedded DL denoising, cardiac motion estimation and correction; and T4: DL reconstruction methods with both RMC and CMC, with projection data binned using respiratory surrogate signals derived in SA1. SA3: Perform multicenter clinical reader studies (6 clinicians, 3 institutions) to validate the new algorithms and compare to current clinically-available methods based on diagnostic performance and repeatability in assessing both perfusion and wall motion defects. T1: In comparison to baseline clinical reconstruction, evaluate added benefit of: a) including attenuation and scatter correction, and b) additionally including RMC; T2: Validate DL for improvement of perfusion and function (wall motion) task performance at full-count levels; and T3: Validate DL for improvement of task performance at reduced counts.