Molecular Imaging & Translational Research, University of Tennessee Graduate School of Medicine, Knoxville, TN, United States of America.
The University of Tennessee: Electrical Engineering and Computer Science, Knoxville, TN, United States of America.
PLoS One. 2022 Aug 31;17(8):e0272768. doi: 10.1371/journal.pone.0272768. eCollection 2022.
Positron emission tomography (PET) is susceptible to patient movement during a scan. Head motion is a continuing problem for brain PET imaging and diagnostic assessments. Physical head restraints and external motion tracking systems are most commonly used to address to this issue. Data-driven methods offer substantial advantages, such as retroactive processing but typically require manual interaction for robustness. In this work, we introduce a time-of-flight (TOF) weighted positron emission particle tracking (PEPT) algorithm that facilitates fully automated, data-driven head motion detection and subsequent automated correction of the raw listmode data.
We used our previously published TOF-PEPT algorithm Dustin Osborne et al. (2017), Tasmia Rahman Tumpa et al., Tasmia Rahman Tumpa et al. (2021) to automatically identify frames where the patient was near-motionless. The first such static frame was used as a reference to which subsequent static frames were registered. The underlying rigid transformations were estimated using weak radioactive point sources placed on radiolucent glasses worn by the patient. Correction of raw event data were achieved by tracking the point sources in the listmode data which was then repositioned to allow reconstruction of a single image. To create a "gold standard" for comparison purposes, frame-by-frame image registration based correction was implemented. The original listmode data was used to reconstruct an image for each static frame detected by our algorithm and then applying manual landmark registration and external software to merge these into a single image.
We report on five patient studies. The TOF-PEPT algorithm was configured to detect motion using a 500 ms window. Our event-based correction produced images that were visually free of motion artifacts. Comparison of our algorithm to a frame-based image registration approach produced results that were nearly indistinguishable. Quantitatively, Jaccard similarity indices were found to be in the range of 85-98% for the former and 84-98% for the latter when comparing the static frame images with the reference frame counterparts.
We have presented a fully automated data-driven method for motion detection and correction of raw listmode data. Easy to implement, the approach achieved high temporal resolution and reliable performance for head motion correction. Our methodology provides a mechanism by which patient motion incurred during imaging can be assessed and corrected post hoc.
正电子发射断层扫描（PET）在扫描过程中容易受到患者运动的影响。头部运动是脑 PET 成像和诊断评估的一个持续问题。通常使用物理头部约束和外部运动跟踪系统来解决这个问题。数据驱动方法具有很大的优势，例如回溯处理，但通常需要手动交互来保证稳健性。在这项工作中，我们引入了一种基于飞行时间（TOF）的正电子发射粒子跟踪（PEPT）算法，该算法可以方便地实现完全自动化的数据驱动的头部运动检测，以及对原始列表模式数据的自动校正。
我们使用了我们之前发表的基于 TOF-PEPT 的算法（Dustin Osborne 等人，2017 年；Tasmia Rahman Tumpa 等人，2021 年）来自动识别患者接近静止的帧。第一个这样的静态帧被用作参考，随后的静态帧被注册到该参考帧上。使用放置在患者佩戴的透明玻璃上的弱放射性点源来估计基本的刚性变换。通过在列表模式数据中跟踪点源来实现原始事件数据的校正，然后对点源进行重新定位，以允许重建单个图像。为了进行比较目的，我们实现了基于帧的图像配准校正。使用原始列表模式数据为我们算法检测到的每个静态帧重建一个图像，然后应用手动地标注册和外部软件将这些图像合并为一个单一的图像。
我们报告了五项患者研究。TOF-PEPT 算法被配置为使用 500 毫秒的窗口来检测运动。我们的基于事件的校正产生的图像在视觉上没有运动伪影。我们的算法与基于帧的图像配准方法的比较结果几乎无法区分。定量分析发现，当将静态帧图像与参考帧图像进行比较时，前者的 Jaccard 相似性指数在 85-98%之间，后者的 Jaccard 相似性指数在 84-98%之间。
我们提出了一种完全自动化的数据驱动方法，用于运动检测和原始列表模式数据的校正。该方法易于实现，实现了高时间分辨率和可靠的头部运动校正性能。我们的方法提供了一种机制，可以在成像过程中评估和事后校正患者运动。
