A preliminary evaluation of a high temporal resolution data-driven motion correction algorithm for rubidium-82 on a SiPM PET-CT system

Abstract

Background

In myocardial perfusion PET, images are acquired during vasodilator stress, increasing the likelihood of intra-frame motion blurring of the heart in reconstructed static images to assess relative perfusion. This work evaluated a prototype data-driven motion correction (DDMC) algorithm designed specifically for cardiac PET.

Methods

A cardiac torso phantom, with a solid defect, was scanned stationary and being manually pulled to-and-fro in the axial direction with a random motion. Non-motion-corrected (NMC) and DDMC images were reconstructed. Total perfusion deficit was measured in the defect and profiles through the cardiac insert were defined. In addition, 46 static perfusion images from 36 rubidium-82 MPI patients were selected based upon a perception of motion blurring in the images. NMC and DDMC images were reconstructed, blinded, and scored on image quality and perceived motion.

Results

Phantom data demonstrated near-perfect recovery of myocardial wall visualization and defect quantification with DDMC compared with the stationary phantom. Quality of clinical images was NMC: 10 non-diagnostic, 31 adequate, and 5 good; DDMC images: 0 non-diagnostic, 6 adequate, and 40 good.

Conclusion

The DDMC algorithm shows great promise in rubidium MPI PET with substantial improvements in image quality and the potential to salvage images considered non-diagnostic due to significant motion blurring.

Introduction

Unlike SPECT, in myocardial perfusion PET images are acquired throughout infusion of tracer and vasodilator stressing agents. This increases the likelihood of motion of the heart during the image acquisition introducing a varying degree of non-uniform blurring in the reconstructed images, hampering image interpretation and, in extreme cases, rendering the images either non-diagnostic or leading to incorrect interpretation.1, 2, 3, 4 Motion remains an ever-present challenge in myocardial perfusion imaging. In rest-stress myocardial perfusion PET, the principal aim is to ensure that the static relative perfusion images preserve diagnostic image quality and it is these images that are arguably the most susceptible to this motion blurring, as a result of being averaged over several minutes.

We consider that there are three types of motion that can degrade static perfusion images: cardiac contraction; periodic motion, usually associated with regular breathing motion; and non-periodic motion. Non-periodic motion can be further divided into two sub-types: a gradual baseline drift of the heart position (sometimes referred to as “cardiac creep”) and irregular oscillation due to irregular breathing motion and movement as a result of discomfort arising from the stress agent side effects.⁵ At least one commercial software package is available which addresses the issue of cardiac contraction blurring to produce “Motion Frozen” static images.^6,7 This only addresses cardiac contraction and so additional corrections are required to address the second and third types of heart motion discussed here. The impact of respiratory motion correction has been a focus of PET for many years—both in oncology and cardiology.^8,9 External hardware options, such as optical systems or expandable belts, are available to estimate internal motion from external measurements from the patient and produce a finite number of respiratory gated images.¹⁰ These respiratory gates can be defined by either temporal (phase) or displacement (amplitude) methods, where image gates represent either equal time periods between end inspiration and end expiration or the position of an organ between end inspiration and end expiration, respectively. One method to utilize these respiratory gated images to perform motion correction in cardiac PET is to simply produce images of the myocardium from one respiratory gate or in the quiescent motion phase.^11,12 However, this approach inevitably leads to inferior count statistics due to the division of acquired data into multiple gated images.^9,13,14 Alternatively, it is possible to shift the position of the heart in each gate to align with one particular reference gate to retain the total number of counts.^15,16 However, it has been suggested that this post-reconstruction alignment produces inferior image quality compared with inserting the motion information directly into the reconstruction process.¹⁷

While the use of external devices has been demonstrated to be effective, they require additional setup time and precise synchronization with the PET acquisition. They also exhibit additional technical challenges that may result in decreased robustness in a clinical setting.¹⁸ Consequently, several studies have investigated the use of data-driven respiratory motion tracking and correction in cardiac PET.^11,19,20 However, these cases have evaluated data with either fluorine-18 FDG or nitrogen-13 ammonia but not rubidium-82, which has a substantially shorter half-life. The rapidly changing tracer kinetics of rubidium-82 may also provide an additional challenge²¹ that has not yet been exhaustedly addressed in the literature.

These aforementioned techniques to remove respiratory motion—either from external trigger or data-driven methods—are commonly based around the formation of a series of respiratory gated images of either fixed temporal or spatial width.^12,15 While these methods may work when there is regular periodic breathing motion with a static baseline, they may not be the optimum technique for non-periodic motion in pharmacological stress-induced cardiac imaging.^13,16 Two recent publications have highlighted the occurrence of non-periodic motion with a drift in baseline position in dynamic rubidium-82 reconstructions for calculation of myocardial blood flow (MBF).^22,23

Current studies addressing motion in rubidium cardiac imaging have focused on the impact of inter-frame motion present in the dynamic reconstructions for MBF calculation.^22,24, 25, 26 However in these cases, the error arises from incorrect time-activity curves due to segmentation errors. Simply applying frame-by-frame motion correction to the PET data by manual translation is usually sufficient to significantly reduce these errors and this functionality is available on some commercial software packages. Despite the value of motion correction in dynamic frames, there is very little work evaluating motion correction of static perfusion images. This is likely to be due to the difference in correction: dynamic motion correction is inter-frame translation, which is relatively straightforward, whereas motion correction of a static image is inherently intra-frame.

Frame-by-frame correction has been demonstrated in a single case study by Thompson et al. to be beneficial for static perfusion images.⁴ In this study, the authors reconstructed a non-diagnostic static perfusion image as a dynamic image of thirty 10-second frames. The individual frames were then aligned using rigid translations and summed to form a motion-corrected static image. Similar techniques have been recently presented from our institution.² However, this was found to be impractical as reconstructing numerous multiple frames is labor and resource intensive and not considered practical for a center with high throughput. In addition, this frame-by-frame technique does not correct for motion that is high frequency to the degree that significant intra-frame motion still occurs in these short frames. The solution is to reconstruct a greater number of shorter frames to improve temporal sampling but this worsens image quality in each frame making visual alignment challenging and also heightens the burden on the reconstruction system.

The final aspect to consider in this study is the improvements from developments in scanner technology. Our institution uses a Biograph Vision 600 PET-CT scanner (Siemens Medical Solutions USA, Inc.) that employs Silicon Photomultipliers (SiPM) for signal detection that offer a significant increase of performance over systems that use traditional photomultipliers. The improved Time-of-Flight (TOF) performance, sensitivity and spatial resolution produces a notable improvement to image definition for rubidium-82 images in spite of the appreciable positron range of rubidium-82 (approximately 5 mm in soft tissue).²⁷ Having performed nearly 2000 rest-stress rubidium scans on the Biograph Vision, the degrading impact of motion on the static perfusion images is more apparent and hence the correction for this becomes ever more desired.

In this work, we present preliminary findings from a prototype Data-Driven Motion Correction (DDMC) algorithm that has been specifically designed for cardiac PET applications. The algorithm acts on listmode data prior to reconstruction and hence eliminating the need to reconstruct and align multiple short images. We leverage the improved TOF performance, sensitivity, and spatial resolution possible with a modern SiPM PET system to perform motion correction at a temporal resolution that has not previously been possible. The algorithm has been applied to phantom and patient studies.