Fast deformable contour propagation for intra-fraction adaptive MR-guided prostate radiotherapy

Purpose or Objective
Intra-fraction (IF) adaptation and full re-planning for ultra-hypofractionated MR-guided prostate cancer (PCa) radiotherapy would allow for offering a completely non-invasive treatment, i.e., without a need for fiducial markers, rectal spacers, and urinary catheters. However, progression towards IF re-planning workflows using repetitive MR imaging is hindered by, among other things, a need for extensive online manual adaptation of the propagated contours. A fast and accurate auto-contouring solution that reduces the need for manual adaptation is needed. Here we assess the clinical usability of IF propagated contours by a deformable image registration (DIR) algorithm.

Materials and Methods
Ten PCa patients treated with 5x7.25 Gy on a 1.5T MR-Linac were included (50 fractions in total). During each fraction multiple (≥2) T2-weighted MR scans were acquired (110 scans in total, 3 min scan time). A normalized gradient field-based DIR algorithm (EVolution, Denis de Senneville, Phys Med Biol. 2016 Oct 21;61(20):7377-7396) was used to sequentially register MR scans (MR1-MR2, MR2-MR3, etc.) for each fraction and to propagate the contours (CTV, bladder, rectum). The CTV included the prostate body, GTV with a 4 mm margin, and up to 1/3rd of the seminal vesicles. The propagated contours were judged by a physician on clinical usability using two criteria. First, for each structure, the need for adaptations within a 2 cm ring (1 cm in craniocaudal direction) around the CTV was assessed on a four-point scale, ranging from none to multiple major adaptations. Second, it was judged if manual adaption of all structures combined could be executed within 3 min, to allow for fast intra-fraction adaptation. Results were stratified by the interval between the MR scans (<10 min versus ≥10 min, ‘short’ and ‘long’), to represent shorter and longer intervals for IF adaptation and re-planning.

Results
Mean (SD) interval between scans was 5.7 (±1.4) min ('short', n=10 pairs) and 20.0 (±5.2) min ('long', n=50 pairs). Table 1 shows the need for adaptations. In 4 (40%) 'short' and 9 (18%) 'long' cases none of the contours needed manual adaptation. All 'short' interval cases could be adapted within 3 min. For the 'long' interval, 3 min was sufficient in 46 (92%) cases. The remaining four showed large IF rectal deformations, in addition to a need for CTV and/or bladder adjustment.

Conclusion
EVolution performed very well for IF contour propagation of bladder and prostate CTV. Overall, rectum contours are acceptable but sometimes need more (extensive) adaptations to perfectly fit the anatomy. Nevertheless, adaptation times were <3 min for the majority of 'long' interval cases and all 'short' interval cases, indicating feasibility for an IF adaptive workflow using repetitive MR imaging and multiple re-planning steps. For cases with extreme IF rectal deformations, 3 min might not suffice. While EVolution provides acceptable results, our future work will include additional algorithms to find the optimal approach.