In high-dose-rate (HDR) prostate brachytherapy, a high-intensity radiation source (e.g. iridium: Ir-192) is inserted into the tumor temporarily. Since a high dose is given in a single fraction, MR guidance is important for safe dose delivery. We are developing a robotic MR-guided HDR brachytherapy procedure [1], aiming at intraprocedural real-time tracking of the HDR source in vivo. During treatment, the HDR source is stepping through a catheter or needle to stay a predetermined time (minimum typically 0.3s) at each of a predefined number of irradiation positions. The artifacts generated by the magnetic susceptibility of the paramagnetic materials of an HDR source, can be exploited to localize the source by MR artifact simulation and phase-only cross correlation [2]. However, to be able to detect the HDR source positions for dose verification, fast imaging and post-processing (<0.3s) are needed. The purpose of this work was to increase the temporal resolution for localization of an HDR source in real-time, by introducing two techniques to accelerate: use of lower-resolution images (i.e. faster imaging) with subpixel source localization and the use of parallel imaging.

MR acquisition: Measurements were done using a non-active iridium source (Flexitron, Elekta) placed inside a tube, fixed in an agar phantom. MR imaging was performed on a 1.5T scanner (Ingenia, Philips, Best, The Netherlands), using a 2D dynamic FFE sequence. For the purpose of higher temporal resolutions, spatial resolutions were varied between 1x1mm², 2x2mm², 2.5x2.5mm² and 3x3mm², see Table 1 for scan parameters. Scanning was performed without parallel imaging and by using SENSE=2. To assess the variation in determined source position coordinates, 100 dynamics were used. To assess the ability and accuracy of subpixel localization, artificial subpixel shifts of the image were applied. This was done by performing subsequent MR acquisitions (10 dynamics), while shifting the FOV such that the off center changed in steps of exactly 0.1 mm in either AP-direction or FH-direction.

Simulations: Simulations and post-processing were performed in Matlab (The MathWorks, USA). The source geometry was modeled as a cylinder of iridium (Δχ=47.1ppm, Ø=0.6mm, length=3.55mm), surrounded by a steel capsule (Δχ=15x10³ppm, Ø=0.9mm, length=4.4mm) at a steel cable (Δχ=15x10³ppm, Ø=0.5mm). Based on the susceptibility differences and echo times, the MR artifacts (complex data) were simulated as described in [2].

Post-processing: An overview of the post-processing operations is given in Fig.1. Phase-only cross correlation (POCC) was performed between the experimental image I₁(x,y) and the simulated image I₂(x,y) with the simulated object at its center, calculated in k-space [3]:

$$POCC(k_{x},k_{y})=\frac{I_{1}(k_{x},k_{y}){\cdot}I_2^*(k_{x},k_{y})}{|I_{1}(k_{x},k_{y}){\cdot}I_2^*(k_{x},k_{y})|}.$$

To get subpixel accuracy, Fourier interpolation was applied by zero-padding the POCC results in k-space, with factors such that the resulting spatial resolution in image domain would be 0.1x0.1mm². Finally, inverse Fourier transformation was performed: $$$POCC(x,y)={\tt{FFT}}^{-1}\left[POCC(k_{x},k_{y})\right]$$$, resulting in a POCC image with a peak value at the source position.

HDR source positions could be accurately determined on a 0.1mm-scale in all investigated cases. Fig.2 shows the detected source positions for the cases with varying resolutions, showing that even with 3x3mm² spatial resolution the position could be determined. Repeated imaging and POCC resulted in coordinates with maximum variations of ±0.2 mm, with standard deviations between 0.04 and 0.07 mm in both AP- and FH-directions for all investigated cases (i.e. variations in resolutions, without/with SENSE). Irrespectively of the use of lower-resolutions and SENSE, subpixel image shifts of 0.1 mm could be detected, see Fig.3 for confirmation. Excellent results were obtained independently of the spatial resolutions, proving the robustness of subpixel source localization. Thus, both investigated acceleration techniques have little influence on the ability to accurately determine the source position, enabling imaging with high temporal resolutions. The use of lower-resolution (2x2mm²) images in combination with SENSE=2 resulted in fast imaging with a dynamic scan time of 191 ms for a realistic FOV. Post-processing took <0.15s in Matlab, which can be accelerated by at least a factor 10 using dedicated programming. As such, the complete process of image acquisition, reconstruction and post-processing will be fast enough to localize the source in real-time. If possibly even faster imaging is needed, spatial resolutions of up to 3x3mm² could be used. Future work should aim at confirmation of the presented findings in a less homogeneous phantom, better resembling in vivo conditions.

Fast image acquisition and post-processing can be performed by using lower-resolution images in combination with subpixel source localization and by using parallel imaging, enabling fully automatic real-time position verification of an HDR brachytherapy source. High accuracy and precision (submillimeter scale) were achieved. Therefore, the technique is highly valuable for MR guidance during the robotic HDR brachytherapy treatment.