Today, 80–85% of patients with pancreatic cancer present with advanced unresectable disease.¹ Magnetic Resonance Imaging-guided High Intensity Focused Ultrasound (MR-HIFU) has recently been suggested to provide non-invasive alternative treatment options for pancreatic cancer patients, with a small study reporting positive pain palliation and tumor control effects.² A key advantage of MR-HIFU is feedback control of energy delivery using MR temperature mapping. In aqueous tissues like the pancreas, phase subtraction-based PRFS MR Thermometry ³ is the method of choice, but to successfully apply this method in the pancreas we need to deal with (respiratory) motion. For other moving organs, e.g. the liver, solutions have been proposed to this end. ⁴ It is tempting to transfer methods from one organ to another. However, this is not necessarily straightforward due to some specific properties of the pancreas, such as its size and shape, unpredictable peristaltic motion, presence of major vessels and air cavities in its surroundings.

The purpose of this study was to evaluate how three potential approaches (gating, multibaseline and referenceless ^5-8) perform in the pancreas.

Image Acquisition: Experiments were performed on a clinical 1.5-T MRI scanner (Achieva, Philips Healthcare, The Netherlands) equipped with a clinical MR-HIFU system. Five healthy volunteers were positioned prone after drinking 500mL of pineapple juice to fill the duodenum and stomach. A 3D fat suppressed T1-weighted scan was acquired in breathhold for planning. Volunteers were asked to breath according to an acoustic signal to sustain a stable breathing pattern. For temperature mapping, a dynamic gradient echo series consisting of 3 coronal slices covering the pancreas head was acquired. Scan parameters included: TR = 100 ms, TE = 19 ms, FA = 21, reconstructed voxel size = 2.5x2.5x8 mm3, dynamic scan time = 400 ms, 500 dynamics. Image analysis: Datasets were analyzed offline using three techniques, viz. gating, multibaseline, and referenceless.

Gating: Phase images were selected based on the respiratory belt signal. Expiration images were used to compute temperature change maps, ^3,5 using the first selected image as a reference for phase subtraction.

For motion correction, optical flow based registration was used for multibaseline and referenceless. ⁹

Multibaseline: Prior to the actual thermometry, in a learning stage, a look-up table containing a set of N reference magnitude and phase images (N=40) was acquired. To address susceptibility related phase changes, a pixel-wise linear relation between motion state and registered phase variations was assumed. For this purpose, Principal Component Analysis was employed. ⁶ During the thermometry stage, for each image, the estimated motion pattern was used to obtain a synthetic reference phase map, which was used for phase subtraction.

Referenceless: A synthetic phase background was reconstructed within a circular region in the pancreas head using the 2D near harmonic approach ⁸ (Figure 1A). This computed phase was then subtracted from the measured phase to obtain temperature estimates.

Statistical analysis: For each technique, the temperature standard deviation (TSD) over time was calculated on a voxel-by-voxel basis. For gating and multibaseline the whole pancreas was semi automatically segmented on the T1 image and used as ROI for statistical analysis (Figure 1B). For referenceless, only the cross section of the circular ROI for background reconstruction and the segmented pancreas was used.

All three motion correction methods in the pancreas clearly improve the TSD in every volunteer (Figure 2). Overall, three techniques (gating, multibaseline, referenceless 30mm) push the median TSD under 2.5°C. The observed TSD distribution with gating and multibaseline were similar, medians were between 1.3 to 2.5 °C and the interquartile (IQR) range was between 0.4 to 1.75 °C. For a small area (30mm circle), the referenceless technique performs better, with the median TSD comprised between 1 to 1.4°C and IQR between 0.5 and 1.9°C. However, using a bigger circle close to the edge of the organ, led to a considerably decreased performance level.

PRFS-based temperature mapping in the pancreas seems feasible. Our results suggest that ablation with typical temperature elevations of about 20-30°C; can be monitored using these methods. However, the referenceless can only be applied when the area heated stays within a small region of the pancreas which makes it very case specific and dependent on the heating protocol and conditions.

When the techniques are to be used for monitoring and control of MR-HIFU induced hyperthermia, where small temperature increases need to be measured over extended time periods, temporal filtering may be employed to reduce the noise influence. In addition, the magnetic field drift effect in this particular anatomical area will have to be addressed.