Radiologists and scientists working on MR thermometry.

Microwave ablation is a minimally invasive thermal therapy to kill tumor cells in various human regions. MR temperature imaging using proton resonance frequency (PRF) shift technique ¹ allows us to monitor temperature during the treatment. To ensure patient safety, acceleration of data acquisition is required to obtain temperature images with high temporal resolution. Besides, techniques reducing motion artifacts are also needed especially in abdominal organs. K-space method ² based on hybrid model ³ is an effective solution to these issues, but this approach becomes inaccurate when motion states are not covered by the baseline library and its reconstruction time is relatively long for real time temperature mapping.

In this study, we developed an accelerated temperature estimation method with relatively short reconstruction time and high temperature accuracy in presence of the motion not covered by baseline library.

Theory: The proposed method estimates temperature changes from undersampled k-space based on the hybrid model ³. Different from the conventional k-space method ², several lines in the center of k-space are acquired in addition to the uniform undersampling acquisition to get low resolution images. These low resolution images are used to match the baseline images and eliminate phase shifts due to magnetic field drift. In this way, we can avoid the loop iterations of the conventional k-space method in order to speed up the estimation, and reduce the steps to adjust the sparsity coefficient during the process of reconstruction. Rigid registration is also implemented to further correct temperature errors due to the displacement of the subject by utilizing the center k-space data. The main steps of the proposed method are shown in Fig. 1.

Experiments: A phantom simulation was performed to simulate a Gaussian-shaped heat-induced phase shift (maximum temperature change: 30℃, standard deviation: 2.5 pixels) on the heating target.

To evaluate the ability of the proposed method to estimate temperature when the motion states are not covered by the baseline library, motion simulation was performed to simulate a moving process of the object. The simulated baseline images consisted of 11 images in which the object was moved 2 pixels/frame along the frequency encoding direction from -10 pixels to 10 pixels relative to the starting point. The moving process during the treatment was then simulated in which the object was moved at a speed of 2 pixels/frame from 10 pixels to -10 pixels in the first 12 frames and moved 1 pixel/frame back to 10 pixels in the last 20 frames as shown in Fig. 3c. So the first twelve positions were all included in the baseline library, while half of the last twenty positions were not.

To verify the ability of the proposed method to estimate temperature changes in practice, a phantom heating experiment was performed using an MR-compatible microwave probe (Nanjing ECO Microwave System Co., Ltd, Nanjing, China). A fiber optic thermometer was inserted into the phantom next to the microwave heating probe.

In-vivo non-heated experiment was also performed to acquire spinal cord images on healthy volunteers with free-breathing. The standard deviations of different time points were calculated for comparison.

Fully sampled k-space data were acquired with the 2D FFE sequence on a 3.0T Phillips scanner (Philips, Best, the Netherlands) and then retrospectively under-sampled. Reconstructed temperature maps of the proposed method were compared with SPIRiT ⁴ and the conventional k-space estimation reconstruction thermometry ² using the same undersampling pattern.

The results of phantom heating simulation are shown in Fig. 2. The proposed method results in the lowest RMSEs and uses much shorter time than the conventional k-space method. Fig. 3 shows the results of motion simulation. There are smaller motion-induced temperature errors using the proposed method. The results of phantom heating and in-vivo experiment are shown in Fig. 4 and Fig. 5. The temperature evolution curve of the proposed method agrees well with the optical fiber (Fig. 4a) and the spatial standard deviations of the proposed method are much less than the conventional k-space method in in-vivo data (Fig. 5).Both the phantom heating simulation and the phantom heating experiment demonstrate the ability of the proposed method to estimate temperature with less computation time and higher accuracy. Motion simulation and in-vivo experiment further show the feasibility of the proposed method for monitoring temperature in presence of motion.The proposed method can accelerate MR thermometry at an acceleration factor of 4 and provide temperature maps with improved accuracy and robustness to motion, which makes real time imaging for MR-guided microwave ablation possible.