Soft-tissue based on-line prostate motion assessment in 4D Cine-MR for MR-Linac treatments

Keizer, D. De Muinck, Kerkmeijer, L. G. W., Maspero, M., van Zyp, J. R. N. Van der Voort, Van den Berg, C. A. T., Raaymakers, B. W., Lagendijk, J. J. W., De Boer, H. C. J.

Radiotherapy & Oncology 133 p. S211-S212


Purpose or Objective We develop real-time MR-guided extremely hypofractionated (HF) prostate radiotherapy (RT) with active correction for prostate intrafraction motion. We have collected an extensive 4D cine-MR dataset to study the intrafraction motion of the prostate during the period of a RT fraction. Previously, we have presented a method for accurate automatic prostate tracking based on fiducial gold markers (ESTRO 37). Now we present a method for soft-tissue contrast based tracking that obleviates the need for fiducial markers on an MR-Linac. Material and Methods Thirty patients undergoing HF prostate RT had repeated cine-MR imaging sessions after each of five weekly fractions in a multicenter Medical Ethics board approved study. Each cine-MR session consisted of 55 sequentially obtained 3D datasets (‘dynamics’), acquired with a balanced 3D gradient echo sequence and a voxel spacing of 0.96x0.96x2mm3. Each dynamic was acquired over an 11 second period, with the cine-MR session covering a 10 minute period, similar to the duration of a RT fraction. A clinician delineated the prostate on the first dynamic from which in‐house developed Python code performed soft‐ tissue (ST) tracking of the prostate in subsequent dynamics using a mutual information metric and rigid transformations. We validated the performance of the ST algorithm with previously obtained, ground truth marker tracking (MT) data of the same dataset. Results The algorithm was applied to 7645 dynamics from 139 sessions with a mean processing time of 5.47±0.77 sec (mean±stdev) per dynamic. The success rate (difference between MT and ST result < 1 mm) was 98.93%. We found group translations after 10 minutes of 0.05±0.81mm for X (LR) , 0.83±1.90mm for Y (AP), ‐0.90±1.85mm for Z (CC) and corresponding rotations of ‐0.49±2.13° about X, 0.09±0.58° Y and 0.09±0.73° for Z. After 10 min, 12% of all sessions had demonstrated a 3D displacement > 5 mm. Linear regression and Pearson correlation analysis indicated a good correlation and non‐significant difference with all p‐values < 1e‐5 between the ST and MT algorithm in the X (R=0.934), Y (R=0.966) and Z (R=0.953) directions as shown in Fig. 1. An overview of the ST intrafraction visualization tool is provided in Fig. 2, showing the rotation, translation and full 3D segmentation of the prostate at a specific time point. Conclusion We have developed a fast, robust and accurate ST tracking algorithm in cine‐MR data which was validated against marker tracking. The presented method for soft‐tissue contrast based tracking obleviates the need for surgically implanted fiducial markers during MR‐guided prostate RT on an MR‐Linac.