Clinical implementation of MRI-only radiotherapy treatment workflow for prostate cancer with a standard linac

BACKGROUND AND AIM The purpose of this work is the clinical implementation of radiotherapy treatment workflow for prostate cancer based only on MRI images (MRI-only) with a standard linac, using the technological tools available in a big general hospital. In the recent years the growing interest in using solely MRI images, due to the excellent soft tissue contrast that makes it unique, both for identification and precise delineation of target volumes and organs at risk (OAR), emerged strongly. Especially in those regions characterized by soft tissues, as the pelvis site, the method has several benefits, including improvement of inter-observer robustness in target delineation, a reduction in contoured structures which potentially leads to smaller irradiated volumes and a major efficiency of the workflow by both reducing the time and costs of patients imaging and removing residual registration error compared to a CT/MR workflow. However, there are several challenges in the implementation and application of MRI-only workflow; the most important one is the Electron Densities (ED) estimation. In fact, unlike CT images, where the voxels intensities are directly linked to the physical properties of the tissues ED, required by the Treatment Planning System (TPS) for dose calculation, MRI signals do not depend on ED but correlate with tissue proton density and thus cannot be directly used. Therefore, it is mandatory to convert MRI data to ED maps to allow the dose distribution even in this case. These generated maps are known as synthetic CT (sCT). Several aspects of the implementation have been examined in this thesis in order to validate the overall workflow. MATERIAL AND METHODS Imaging acquisition was performed using a 3D T1 VIBE Dixon gradient echo sequence that is a good compromise between image quality and acquisition time; it takes 2 minutes and 30 seconds and provides 4 contrasts (Water, Fat, In-phase and Out-of-phase) useful for the visualization of target and different OARs. The acquisition on the MR scanner was possible thanks to the use of a home-made table-top, commercial MR safe markers and properly controlled internal laser system; furthermore, the geometric distortions in obtained images were evaluated. A hybrid method was adopted to generate sCT from MRI; it consists in a combination of bulk ED and multi atlas-based approach. For this purpose, the MR images of 20 volunteers and patients undergoing diagnostic acquisitions were collected, manually contoured in bones and main OARs, and then inserted, after verification by a radiation oncologist, as reference in the database of ADMIRE software (research version 1.13.5, Elekta AB, Stockholm, Sweden), utilized for automatic contouring process. The software approximates the anatomy contours by comparing several individual atlases, applying elements of maximum likelihood forms to a new patient image-set, and creates a structure set to fit the actual patient’s anatomy. Therefore, sCTs were finally generated in the TPS by assigning to each structure bulk ED values, calculated averaging over 20 patients who previously underwent prostate radiotherapy. Once the sCT was generated, different evaluations were made using quantitative methods such as dose-volume histogram (DVH) and 2D local gamma analysis, in order to test the dose differences between the sCT and the gold standard CT, both in terms of homogeneous EDs assignment and contours delineated on the MR images. At first, 20 VMAT prostate treatment plans, calculated on standard CT images with 10 MV X-rays, were recalculated using the same field, segments and monitor unit (MU) arrangements on the sCT, where only fixed EDs were assigned instead of the heterogenous ones. Afterwards, the combination between the assignment of an average ED and the different contours in the two imaging modalities was evaluated for 7 patients undergoing both CT and MR acquisitions. The optimized plans, calculated on the CT images, were first copied using the same field, segment and MU arrangements and recalculated on the sCTs, obtained by MRI, and thus were optimized again keeping the same plan constraints as CT-based ones. Finally, the feasibility of patient positioning in the linac room using CBCT-MR matching was verified. For 9 selected images of different fractions and patients, the shifts obtained through CT-CBCT registration were compared to the those resulting from the evaluations of three radiation oncologists, who blindly and manually matched the CBCT images with those from MRI. RESULTS AND DISCUSSIONS The MR sequence used for this work was optimized in order to achieve adequate image quality in a reasonable time, reducing the artifacts and minimizing the patient discomfort. Furthermore, the internal laser system of the MRI scanner, used patient positioning, was checked. It was verified that the laser cross, projected from a single point inside the scanner, was centered in the isocenter and not tilted. The mean tilt angle was 0.3°± 0.2° (range 0.2° to 0.4°) in all positions tested, corresponding to difference approximately of 1±0.4 mm at 15-20 cm from the isocenter. In addition, the geometric distortions, evaluated in the absence of the patient less than 2 mm, are minimized by the active shimming system of the scanner. The auto-contouring delineation process performed by ADMIRE (research version 1.13.5, Elekta AB, Stockholm, Sweden) was executed in an average time of approximatively 20 minutes and considered satisfactory by radiation oncologists; further 5 to 10 minutes were needed to better define some contours and to identify the target. Regarding the evaluation made on the same CT images, where only the EDs were changed, the mean deviation of PTV and OARs DVH parameters were approximately 0.5%, except for the rectum were the percentage difference reached higher values (Dmean of 1% and V40 of 2.4%) due to different filling. In addition, further evaluations performed by 2D local γ-analysis resulted in average passing rate from 98.5% to 99.4% in the three views, using an acceptance criterion of 2%-2mm. On the sCT obtained by MR images, despite the mean percentage differences in PTV coverage were limited, below the 1% in the first case and around 0.5% in the second one, the differences for OARs were considerably higher; mean percentage differences for bladder and femoral heads were approximatively -25% and -2% in both the described steps, while Dmean of rectum ranged from +11.5% to +4% and V40 passed from +26% to +8%. Anyhow, this issue is not relevant in the implementation of the workflow, as only MRI acquisitions will be made for planning purposes. Rather, this behaviour is representative of the differences in filling of different OARs, such as rectum and bladder, between two consecutive fractions of the treatment. Moreover, however, the PTV and OARs compliance with the constraints, used in our center, was evaluated for the re-optimized plans: for 7 patients and 8 different parameters, only 1 time over 56 a constraint exceeded the limits. Finally, as far as the matching between CBCT-MR for patient positioning before each treatment fraction, the averaged differences in displacements respect to standard CT-based method, resulted 2.8±1.7 mm, 3.8±1.6 mm and 0.1±1.4 mm for transverse, longitudinal and vertical directions respectively, despite the physiological differences of the rectum and bladder between the two imaging modalities. CONCLUSIONS This study demonstrates that MRI-only workflow for prostate patients seems to be feasible using the clinical optimized method, providing a better contrast in the structures of clinical interest. Therefore, the future goal is to gradually move towards the use of MRI-only in the clinical routine. Meanwhile, the proposed method will be used for special cases, such as patient with metal hip prosthesis.