Session Item

In the current workflow for prostate radiotherapy the advantage of using MRI for target delineation is reduced due to the uncertainty in the registration with the planning CT. This systematic error contributes to the required PTV margin. In an MR-only workflow a pseudo-CT for treatment planning is generated from a specific MRI sequence, which eliminates the registration error between MRI and planning CT. However, the fiducial markers that are required for position verification are not visible on the pseudo-CT. We have created a semi-automatic method, for burning in fiducial markers on the pseudo-CT.  The goal of this study is to examine inter-observer variation of the resulting marker positions.

Twenty patients that received prostate radiotherapy at our department, were included in this study. After giving informed consent, a mDIXON FFE MRI sequence was added to the standard clinical protocol for the generation of a pseudo-CT. Prior, patients received four 1x5mm gold fiducial prostatemarkers. Seven observers, five radiotherapy technologists (RTT) and two medical physicists, were instructed to delineate areas with no signal on the water-weighted reconstruction of the mDIXON (slice-thickness 2.5mm, in-plane resolution 1.7x1.7mm) where they expected the fiducials to be.  For optimal CBCT-pseudoCT matching, they added a 2mm margin. A BFTE SPAIR sequence, which was part of the clinical protocol, was used to identify the positions of the fiducial markers (Figure 1). For each delineation the center of mass (COM) position was determined of the one, five or ten voxels with the lowest signal, to be used for burning in on the pseudo-CT. For each marker the standard deviation (SD) of the anterior-posterior (AP), left-right (LR) and cranio-caudal (CC) component of the COM position was calculated.

At the time of the examination two patients had three markers, leading to a total of 78 markers. For 94%, 91% and 90% of the markers the SD of the CC component of the COM position was less than 0.5 mm for one, five, and ten burned-in voxels, respectively. For the LR and AP components the SD was less than 0.5 mm for 97% of the markers (Figure 2). The largest SD occurred for the CC component, which is the slice direction, and was 4.5 mm for one voxel and resulted from a misidentified marker. Increasing the number of voxels to five reduced the maximum SD for the CC component to 2.2mm. COM positions of the markers differ most when implanted very near to another, which makes it difficult to distinguish the markers.

The inter-observer  variation of the burned-in fiducial marker positions was small in the in-plane direction and larger in the slice direction. Burning in more voxels reduced the variation in the slice direction considerably. In a clinical workflow, marker delineations should be checked by a second RTT or physicist, to prevent misidentification of makers. Based on the results gathered from burning in five voxels, this method was found suitable for clinical use.