Part I | Chapter 3 54 Materials and methods Patients and imaging data Ten prostate cancer patients treated with 5 x 7.25 Gy on a 1.5 T MR-Linac (Unity, Elekta AB, Stockholm, Sweden) were included (50 fractions). All patients were part of an institutional review board approved registration and imaging study. During each fraction, an initial daily MR (INI) scan and position verification (PV) scan were acquired (Figure S1, Supplementary A). During some fractions, a second PV scan was acquired, i.e. when the time between the PV scan and end of treatment planning was too long due to unforeseen circumstances. In total, 110 MR scans (50 INI, 60 PV) were included (Table S1, Supplementary B). Deformable image registration and contour propagation For each fraction, the INI scan was registered to the PV scan (n = 50) and the contours were propagated from the INI to PV scan (Figure S2, Supplementary C). In case of an additional PV scan (n = 10), the first PV scan was registered to the second PV scan. After registration, the Clinical Target Volume (CTV) and OARs (bladder and rectum) contours were propagated from the prior to the latter scan (Figure 1). The CTV contour included the prostate body, Gross Tumour Volume (GTV) with a 4 mm margin, and up to 1/3rd of the seminal vesicles. DIR and contour propagation were performed using the in-house developed ‘EVolution’ algorithm.19 The algorithm optimises the local alignment between similar contrast patterns within the registered images, making it suitable for both mono- and multi-modal image registration. The algorithm was primarily chosen due to its previously demonstrated clinically acceptable accuracy for contour propagation.18,20 Moreover, the method is highly parallelisable, facilitating a fast convergence of < 2 sec for mono-modal MRI registration (256 x 256 x 128 image size) using the Compute Unified Device Architecture (CUDA) and when performed on a NVidia TITAN V graphics processing unit. Finally, it requires a low number of input parameters, which can be maintained at fixed values over the entire duration of a treatment.18,21 Therefore, the algorithm can be seamlessly integrated into online adaptive workflows. Clinical assessment of contours After DIR and contour propagation, the contours were judged by two independent observers (physicians) on clinical usability using two criteria. First, the need for adaptations within a 2.5 cm ring around the CTV (1.5 cm craniocaudally) was assessed on a four-point scale for each structure separately. The assessment scale ranged from “none” (1) to “multiple major adaptations needed” (4) (Figure S2, Supplementary C). The 2.5 cm ring included the region of interest (high-dose region). Second, it was judged if approving and manual editing of all contours could be executed within 3 min, to allow for sufficiently fast cycle times. Results were stratified by observer and the interval between the sequential MR scans (< 10 min versus ³ 10 min, ‘short’ versus ‘long’) to assess contour quality for shorter and longer MR scan intervals.
RkJQdWJsaXNoZXIy MTk4NDMw