The creation of a respiratory motion corrected end-diastolic image (MoCo-4D) minimizes motion and improves image quality compared to dual-gating. This method also saves 25% of additional data compared to dual-gating. The proposed methods could be implemented using existing software on a clinical PET/CT system.
MoCo-4D was superior in majority of the image quality metrics with optimum motion minimization, showing statistically significant improvement compared to both NG and DG.
MoCo-4D showed increased myocardial SUVmean and lower blood pool SUVmean compared to other methods (Table 2, Figure 2). MoCo-4D showed improved SUVmean recovery compared to DG and MoCo and reduced spill-over compared to NG.
In terms of image quality (Table 3, Figure 3), MoCo-4D showed superiority in a majority of measured parameters compared to DG. For CR and SNR, this can be seen clearly.
CV increased in DG, MoCo and MoCo-4D to similar level (~32-33%). For DG, this is caused by increase of noise. As motion correction is performed post-reconstruction, noise is averaged in the corrected image, increasing CV in MoCo and MoCo-4D.
MoCo-4D reduced motion compared to NG, with 16.6 mm vs 20.9 mm MWT (Table 3). MoCo and DG images had similar MWT (16.6-16.8 mm), indicating that no residual blurring was introduced by MoCo. MoCo-4D resulted in slightly smaller MWT subject-wise (Figure 4).
In the visual evaluation, spill-over in the blood pool was reduced with DG and MoCo methods (Figures 5 and 6). Slight increases in local uptake were visible in MoCo methods, with no additional blurring compared to DG. The uptake profiles were also sharper due to motion minimization (Figure 7).
Comparison of 4D-CT and CINE-CT
We compared two methods for attenuation correction. Using CINE-CT in cardiac imaging improves accuracy.8,9 We have shown that motion correction benefits from 4D-CT with the same radiation dose as CINE-CT. However, CINE-CT increases dose and might not be justified, if no benefits are seen over the standard CTAC protocol.18
Comparison to Previous Studies
MoCo-4D has higher MWT (16.6 mm) than 13.8 mm reported in Ref. 5 and 11.6 mm in Ref. 17, which applied cardiac5 and cardiac-respiratory17 motion correction. We used a 6 mm Gaussian filter from our clinical protocol, which is optimal for noise suppression but reduces resolution. A 2 mm filter used in Refs. 11,17 might be more optimal. As in previous studies, we noted increased CR and CNR, although direct comparison is challenging due to different methodologies.
A trade-off exists between motion minimization and noise increase in dual-gating, degrading registration accuracy.6 We applied 5/5 cardiac and respiratory bins according to our clinical protocol versus 8/4 in Ref. 6. This is nearly identical to the optimal amount reported in Ref. 19, although on a different system. A more optimal amount might be discovered with systematic testing, which is out of the scope of this study.
Differences in SUVmean and CV are attributed to VOI delineation, which was performed independently for each image. This causes variability in VOI size and the activity covered by the VOI, increasing standard deviation and CV. We quantified SUVmean instead of SUVmax, since the dependence of SUVmax on counting statistics has been shown20 whereas SUVmean is more robust.21 The method was also evaluated with a small number (N = 13) of patients, prompting follow-up studies with larger patient groups.
Motion correction is challenging in plaque imaging. All subjects in this study had myocardial uptake, thus the motion correction performance with coronary lesions of low uptake was not evaluated. A level set-based registration has been applied in [18F]-NaF PET,22 using coronary CT angiography for PET registration.
Our method saves 39% of the data, as only the diastolic phase is used for motion correction. Thus, a dual motion correction approach as in Ref. 17 might be preferred, which can preserve 100% of the data. Applying cardiac motion correction as in Ref. 23 might also improve image quality and MWT, however, these approaches are not yet implemented on PET/CT systems. Moreover, a triple-motion correction including gross patient motion correction might increase performance.21
Finally, sinogram-based motion correction is advantageous for image quality and motion minimization. A challenge has been adopting sinogram-based methods in clinical routine, due to complexity and user-defined reconstruction, challenging clinical comparison.24 However, one such approach has been recently implemented using vendor-based reconstruction.24 Thus, modifying our method to sinogram-based correction could improve performance.
New Knowledge Gained
We implemented a motion correction method which corrects respiratory motion over the diastolic phase combined with 4D-CT. The method improves image quality and motion minimization qualities significantly and is relatively simple to implement clinically using existing methodologies. We’ve shown the clinical applicability of the proposed approach and that 4D-CT should be used in combination of motion correction for optimum results.