The electrical characteristics of RF coils are affected by the objects they couple to. This has been used to detect respiratory and cardiac motions (1,2). We hypothesize that using vendor provided parallel transmit (PTX) SAR monitoring hardware (Siemens 7T) the reflected power can be monitored on every RF pulse and we can relate this measurement to motion using MR images. This strategy uses an MR image based learning interval to provide a hybrid method to quantify in real-time the diaphragm position at a 4.5ms (image TR) temporal resolution.

The Siemens 7T PTX local SAR monitor samples the forward and reflected power of all RF pulses on each transmit channel via directional couplers that are then digitized by the imager. A programme was written on the image reconstruction computer to calculate the reflection coefficients (S_ii) at the end of each RF pulse on all eight transmit channels. A change in S_ii occurs when the diaphragm changes position, this S_ii change is transformed into a quantitative diaphragm position using a series of MR images to measure the diaphragm position during an initial learning interval.

To assess the method two experiments were setup on a human 7T scanner (Siemens). The first, to assess the accuracy of the diaphragm position, and the second to evaluate whether this operates effectively within a real-time cardiac cine acquisition. Five healthy volunteers were scanned according to our institutions ethical guidelines. A sagittal gradient echo (GRE) acquisition placed over the right hemi-diaphragm with a TR/TE of 4.5/1.9ms, image matrix of 304x156, resolution of 1.0 (HF) x 1.9mm (AP), and flip angle ~2°. The image was repeated 513 times every 321ms for 2min 45s. The eight complex S_ii were calculated from the central 10μs of each RF pulse. The diaphragm edge was measured in each image. The first 20s of measurements were used as a learning cycle by performing a linear regression between the set of S_ii and the diaphragm edge. In the regression the complex S_ii was separated into real and imaginary components. One image, every 8s, was used as an absolute position reference to remove effects of systematic drift in S_ii.

In the second experiment a pulse sequence was constructed to include a diaphragm image GRE navigator (3) that runs repeatedly for 20s at the start of the sequence with three dummy excitation pulses between each navigator. After this the navigator was repeated every 8s to measure and account for systematic drift in S_ii (as illustrated figure 1). The navigator duration was 66ms, resolution of 1.4mm (HF) x 15mm (AP), slice thickness of 15mm, flip angle ~2°. The position of the diaphragm was determined using the method above and used with a 5mm window to prospectively respiratory gate the acquisition. A non-gated acquisition was acquired for comparison. A retrospectively ECG gated cine was acquired. Flip angle 12°, FOV 320x300, resolution 1.4x1.4x8mm³, TR/TE 5.8/3.06 ms, bandwidth of 744Hz/pix, and a temporal resolution of 52ms. A Kalman filter was implemented in both experiments to increase the confidence when an offset is applied.

For experiment 1: figures 2 and 3 show representative traces of the diaphragm position measured with the S_ii method compared to that measured in the images for subjects 3 and 2 respectively, figure 4 lists the mean ± standard deviation of the difference between the S_ii diaphragm position and that measured in the images without filtering. The last column of the table shows the accumulated system drift over the scan.

For experiment 2: figure 4 shows a single cardiac cine frame for non-diaphragm-gated and diaphragm-gated cine acquisitions. There was a delay of 23ms due to processing and networking constraints between measurement of the diaphragm position and updating the RF and gradient waveforms.

For all subjects except the outlier subject 2, the mean±SD of the difference between the image and S_ii prediction was -0.1±1.2mm. For subject 2, who’s trace is shown in figure 3, both the image and S_ii measures indicate a large unusual motion, perhaps coughing at the end of the scan.

This method poses a number of benefits over conventional respiratory monitoring, i) it requires no additional hardware as it uses the existing RF coil and SAR monitoring hardware, ii) the pulse sequence maintains a steady state, iii) diaphragm positions are known with a real-time lag of 23ms

In conclusion, real-time diaphragm navigation has been demonstrated using vendor provided PTX local SAR monitoring hardware complemented by a position learning phase then infrequent navigator images.