TH-EF-BRA-07: Evaluation of Internal Target Volume Derived From a Prototype 4D-MRI Sequence with 3D Radial Stack-Of-Stars Trajectory and K-Space Self-Gating




4D-MRI based on resorting of 2D-cine-MRI images shows great potential to assess tumor motion more accurately compared to 4D-CT, however, it suffers from low through-plane resolution and stitching artifacts. 4D-MRI based on 3D acquisition results in stitching-artifact-free images with high in-plane and through-plane resolutions. In this study, we report our early clinical experience of a prototype 4D-MRI sequence with 3D stack-of-stars (SOS) trajectory for internal target volume (ITV) definition.


Ten patients with 13 total lesions (7 pancreatic, 1 lung with 2 lesions, 1 liver with 3 lesions and 1 esophagus) were recruited. 4D-MRI used in-plane radial and through-plane Cartesian sampling. Two imaging orientations, i.e. axial slab (A) and coronal slab (B), were compared. ITVs were derived from 3-bin (ITV3), 5-bin (ITV5) and 10-bin (ITV10) reconstruction protocols. ITV5 was set as standard and minimum expansion of ITV3 needed to encompass ITV5 was derived. Similarity index was calculated from ITV3 and ITV5 (SI3/5), and ITV10 and ITV5 (SI10/5). Imaging noise was calculated for both method A and B. Wilcoxon-rank-sum test was performed with a p value <0.05 deemed as significant.


No significant difference (p=0.34) was observed from method A and B, indicating a uniform imaging noise distribution from 3D acquisition. Imaging noise and artifacts were visually different from different binning protocols, with more bins resulting in more noise, more artifacts and larger ITV. On average, ITV differs by 7% (1%–19%) comparing ITV3 with ITV5 for the patient cohort. For the pancreas sub-group, ITV differs by 4% (1%–6%). An average of 2.3mm (2mm-3mm) expansion was needed for ITV3 to encompass ITV5. SI10/5 was 0.93±0.03 (mean±σ) and SI3/5 was 0.95±0.03.


4D-MRI with 3D SOS trajectory was evaluated on 10 patients. Significant difference in ITV was observed with different binning protocols. Imaging noise was similar irrespective of the imaging orientations.

This work is partially supported by NIH R03CA173273 and CTSI core voucher award.