Respiration‐based sorting of dynamic MRI to derive representative 4D‐MRI for radiotherapy planningPURPOSE: Current pretreatment, 4D imaging techniques are suboptimal in that they sample breathing motion over a very limited "snap-shot" in time. To potentially address this, the authors have developed a longer-duration MRI and postprocessing technique to derive the average or most-probable state of mobile anatomy and meanwhile capture and convey the observed motion variability. METHODS: Sagittal and coronal multislice, 2D dynamic MRI was acquired in a sequential fashion over extended durations in two abdominal and four lung studies involving healthy volunteers. Two sequences, readily available on a commercial system, were employed. Respiratory interval-correlated, or 4D-MRI, volumes were retrospectively derived using a two-pass approach. In a first pass, a respiratory trace acquired simultaneous with imaging was processed and slice stacking was used to derive a set of MRI volumes, each representing an equal time or proportion of respiration. Herein, all raw 2D frames mapping to the given respiratory interval, per slice location, were averaged. In a second-pass, this prior reconstruction provided a set of template images and a similarity metric was employed to discern the subset of best-matching raw 2D frames for secondary averaging (per slice location and respiratory interval). Breathing variability (per respiratory interval and slice location) was depicted by computing both a maximum intensity projection as well as a pixelwise standard deviation image. RESULTS: These methods were successfully demonstrated in both the lung and abdomen for both applicable sequences, performing reconstructions with ten respiratory intervals. The first-pass (average) resulted in motion-induced blurring, especially for irregular breathing. The authors have demonstrated qualitatively that the second-pass result can mitigate this blurring. CONCLUSIONS: They have presented a novel methodology employing dMRI to derive representative 4D-MRI. This set of techniques are practical in that (1) they employ MRI sequences that are standard across commercial vendors; (2) the 2D imaging planes can be oriented onto an arbitrary axis (e.g., sagittal, coronal, axial[ellipsis (horizontal)]); (3) the image processing techniques are relatively simple. Systematically applying this and similar dMRI-based techniques in patients is a crucial next step to demonstrate efficacy beyond CT-only based practice.
4D tumor centroid tracking using orthogonal 2D dynamic MRI: Implications for radiotherapy planningPURPOSE: Current pretreatment, 4D imaging techniques are suboptimal in that they sample breathing motion over a very limited "snapshot" in time. Heretofore, long-duration, 4D motion characterization for radiotherapy planning, margin optimization, and validation have been impractical for safety reasons, requiring invasive markers imaged under x-ray fluoroscopy. To characterize 3D tumor motion and associated variability over durations more consistent with treatments, the authors have developed a practical dynamic MRI (dMRI) technique employing two orthogonal planes acquired in a continuous, interleaved fashion. METHODS: 2D balanced steady-state free precession MRI was acquired continuously over 9-14 min at approximately 4 Hz in three healthy volunteers using a commercial 1.5 T system; alternating orthogonal imaging planes (sagittal, coronal, sagittal, etc.) were employed. The 2D in-plane pixel resolution was 2 × 2 mm(2) with a 5 mm slice profile. Simultaneous with image acquisition, the authors monitored a 1D surrogate respiratory signal using a device available with the MRI system. 2D template matching-based anatomic feature registration, or tracking, was performed independently in each orientation. 4D feature tracking at the raw frame rate was derived using spline interpolation. RESULTS: Tracking vascular features in the lung for two volunteers and pancreatic features in one volunteer, the authors have successfully demonstrated this method. Registration error, defined here as the difference between the sagittal and coronal tracking result in the SI direction, ranged from 0.7 to 1.6 mm (1σ) which was less than the acquired image resolution. Although the healthy volunteers were instructed to relax and breathe normally, significantly variable respiration was observed. To demonstrate potential applications of this technique, the authors subsequently explored the intrafraction stability of hypothetical tumoral internal target volumes and 3D spatial probability distribution functions. The surrogate respiratory information allowed the authors to show how this technique can be used to study correlations between internal and external (surrogate) information over these prolonged durations. However, compared against the gold standard of the time stamps in the dMRI frames, the temporal synchronization of the surrogate 1D respiratory information was shown to be likely unreliable. CONCLUSIONS: The authors have established viability of a novel and practical pretreatment, 4D tumor centroid tracking method employing a commercially available dynamic MRI sequence. Further developments from the vendor are likely needed to provide a reliably synchronized surrogate 1D respiratory signal, which will likely broaden the utility of this method in the pretreatment radiotherapy planning context.