INTRODUCTION

The slow acquisition speed of MRI can limit its applicability for examination of the chest and abdomen, often leading to motion artifacts caused by physiological movements such as cardiac and respiratory motion (1). External sensors (2) such as respiratory belts have traditionally been used for triggering the acquisition. However, this approach is oftentimes unreliable and tends to suffer from poor sensitivity. Self-navigation (3), on the other hand, extract motion information from k-space center, which introduces a dependency on imaging parameters (e.g. TR, number of slices) and can be too slow for certain types of physiological motion. In this work, we demonstrate an alternative approach which provides respiratory information with high temporal resolution. A reference RF signal (4), termed “pilot tone” (PT), is generated by a small radiofrequency transmitter and received by the MR system during each readout. Because the PT amplitude is modulated by the subject’s breathing pattern, a respiratory signal can be derived. The PT approach was tested using a radial MR sequence on human volunteers, and was compared to conventional methods for motion correction.

METHODS

Experiment setup: All data were acquired on a 3T Prisma system (Siemens Healthcare, Erlangen, Germany) using a body coil array. The study protocol included a free-breathing radial stack-of-stars 3D GRE (RAVE) sequence with golden-angle acquisition. It was tested on 15 healthy human volunteers. RAVE imaging parameters included oversampling factor=2, TR/TE=4.0/1.7ms, BW=500Hz/pixel, 240 slices, 800 radial views, 1.6mm isotropic resolution.
Principle: A prototype PT transmitter is a small device placed outside the MR bore with no direct contact with the patient. Physiological movement of the subject causes coil load variations, resulting in modulations of the PT signal detected by the MR receive coil. The PT signal frequency is controlled so that it is imprinted outside the object in image space. To avoid interference between the PT signal and the MR signal (center frequency at 3T = 123.25 MHz), the PT is tuned to transmit ~100 kHz off-resonance. The separation between the imaged object and the pilot-tone (Fig. 1a) is noticeable after FT along the frequency domain. The image (Fig. 1b) and PT (Fig. 1c) information can be separated and a respiratory signal can be extracted by a fit in the frequency domain:$$A\cdot\exp\left(-i\cdot2\pi\cdot\ f\cdot\ t\right)$$ where $$$ f$$$ is the PT frequency and $$$ A$$$ is the complex amplitude signal, which modulates in accordance with the respiration (Fig. 1d). A peak-detection algorithm calculates the PT’s frequency based on the strongest coil and then uses this frequency to calculate the coefficient ($$$A$$$) and its modulation over time for the rest of the coils.
Workflow: During experiments, the PT transmitter is placed outside the MR bore, resulting in a unique signal imprinted in the image data prior to processing (Fig. 2a). The breathing-related PT signal modulation, picked up by each receive coil, is extracted by the peak-detection algorithm and processed by principal component (PCA) analysis, resulting in one respiratory signal from all coils (Fig. 2b). As the next step, the respiratory signal is exploited by an eXtra-Dimensional (XD) reconstruction pipeline (3), which bins the continuously acquired radial data into different respiratory states from inhale to exhale (Fig. 2c).

RESULTS AND DISCUSSION

When comparing respiratory signals from the respiratory belt (Fig. 3a, blue), self-navigation (Fig. 3a, red), and pilot tone (Fig. 3a, green), similar patterns were seen for self-navigation and PT, but the belt signal differed from those patterns. The belt is an analog device that is sensitive to the patient setup, often leading to distorted/clipped signal as experienced here. Similarity between the PT and k-space center signals was noted whether the subject was breathing steadily (Fig. 3b) or unsteadily (Fig. 3c). To reduce dimensionality and noise, PCA analysis, second-order blind identification (SOBI) (5), and coil clustering (6) algorithms were used (Fig. 4a) to process the signals recorded by all MR coils. Respiratory signals based on k-space center (Fig. 4a, pink) missed some respiratory information while PT showed better signal recurrence when processed by the different methods. This result has a direct impact on how motion affects the final images (Fig. 4b), revealing clearer anatomical details like the pancreas and the abdominal wall when using the PT. Moreover, it also influences the image quality along the different respiratory motion states. In a 3D view of another volunteer, use of the PT signal resulted in less motion blurring especially at the liver tip and in the center of liver (Fig. 4c). To further illustrate the reliability of PT, a video of zero-angle acquisitions after Fourier transformation shows how the PT signal (blue) is synchronized with the actual breathing of the subject (Fig. 5).

CONCLUSION

A novel device for tracking respiratory motion, called a pilot tone, showed fewer motion artifacts when compared to conventional k-space self-navigation and respiratory belts. The small dimensions (8 cm) and high sampling rate associated with the PT, offer great potential for tracking breathing signals. Additionally, since the pilot-tone retrieves its information separately from the MR data, it is well suited to tracking motion during the injection of contrast material, which is not possible with k-space-based signals due to spatially varying contrast uptake.

Acknowledgements

We acknowledge support from NIH grant P41 EB0171813 and R01 5R01EB018308.