Session Item

Respiratory motion is a large source of uncertainty for radiotherapy, which can be reduced by treating in the time-weighted average/mid-position (midP) anatomy. A midP image can be derived from a respiratory-sorted 4D-MRI, which usually has a poor resolution. We propose combining a 4D-MRI with an additional, navigator-triggered scan which can be acquired at high resolution, to maximize spatial fidelity and image quality of the midP image. Here, we investigate the quality of improved midP MRI imaging.

We previously developed a simultaneous multi-slice (SMS) accelerated coronal TSE 4D-MRI sequence. A 4D-midP image (2×2×2 mm³ voxel size) can be derived by deformably warping all 4D phases to the midP anatomy, and calculating a time-weighted median (Fig 1a). Combining the 4D-MRI with a navigator-triggered end-exhale axial PROPELLOR (MVXD) scan (Fig 1b), yields a higher-resolution MVXD-midP image (0.5×0.5×3.5 mm³ voxel size). After warping the MVXD image to all 10 4D phases in ADMIRE v3.22.2 (Elekta AB, Stockholm, SWE), the time-weighted average of the deformable vector fields (DVFs) is used to warp the MVXD image to the midP.

Both midP images were calculated for 13 patients and 2 healthy volunteers, scanned on a 1.5T Ingenia MR-sim (Philips Healthcare, Best, NL).

The consistency of the DVFs is quantified using the Distance Discordance Metric (DDM) within the body. The midPs are validated by manual, translation-only registrations of each 4D-phase and midP to end-exhale, based on a reference structure: for 8 patients a tumor (9 tumors in total), and for 2 volunteers the liver-lung interface (6 4D-MRI scans in total). Patient data was used if a clinical GTV delineation was available, the tumor moved due to respiration, and the tumor was visible in the 4D-MRI. The ground-truth translation from the midP to end-exhale, based on all 4D phases, is compared to the translation from the calculated midPs to end-exhale.

Figure 2 shows midP images for an example patient. The MVXD-midP is much sharper than the 4D midP.

Mean (highest 95%) DDM values of 0.9 (3.1) mm (4D-midP) and 1.5 (4.0) mm (MVXD-midP) were found for patients, and 0.9 (3.1) mm and 1.5 (4.0) mm for volunteers. The percentage of DDM values <2 mm was 88% (4D-midP) and 76% (MVXD-midP) for patients, and 87% and 76% for volunteers. Elevated DDM values were found in areas with pulsatile flow (heart, blood vessels) and ghosting (skin), as expected.

The manual registrations show a mean (max) craniocaudal difference between the midP and ground-truth translations of 0.3 (1.0) mm (4D-midP) and 0.5 (1.5) mm (MVXD-midP) for patients, and 0.6 (1.4) mm and 0.8 (1.4) mm for volunteers.

By combining a navigator-triggered MVXD image with an SMS-4D-MRI image, we constructed a dramatically sharper midP image. DDM analysis shows that the MVXD-midP is slightly less reliable than the 4D-midP, but this has limited influence on the regions of interest. Using this higher-resolution MVXD-midP could improve treatment planning.