2528

Comparison of liver motion measured by dynamic MRI and respiration signals obtained by an optical sensor
Julien Sénégas1, Sascha Krueger1, Daniel Wirtz1, Ger Kersten2, Mukul Rocque3, Ivan E. Dimitrov4, Andrea J. Wiethoff5, Keith Hulsey6, Ivan Pedrosa6, and Ananth J. Madhuranthakam6

1Philips Research Laboratories, Hamburg, Germany, 2Philips Innovation Services, Eindhoven, Netherlands, 3Philips Research Laboratories, Eindhoven, Netherlands, 4Philips Healthcare, Gainesville, FL, United States, 5Philips Research North America, Cambridge, MA, United States, 6Department of Radiology and Advanced Imaging Research Center, UT Southwestern Medical Center, Dallas, TX, United States

Synopsis

To monitor the breathing status of a subject and to synchronize data acquisition with respiration, external sensors, such as respiratory bellows, are routinely used. These sensors probe only the local breathing motion, and, hence, their signal quality can vary significantly depending on the individual subject’s physiology and morphology as well as the experience of the MR operator. Recently, optical sensors were proposed as an alternative. The purpose of this work was to compare the signals obtained by an optical sensor and by a pressure-based sensor with respect to their ability to represent the true liver motion during breathing.

Introduction

Coping with respiratory motion is a major challenge in body MRI. To monitor the breathing status of a subject and to synchronize data acquisition with respiration, pressure-based external sensors, such as respiratory bellows, are routinely used. These sensors probe only the local breathing motion, and, hence, the signal quality of such sensors can vary significantly depending on the individual subject’s physiology and morphology as well as the experience of the MR operator. Recently, optical sensors were proposed as an alternative, where video images are processed to extract the displacement of selected body parts induced by breathing1,2. The purpose of this work was to compare the signals obtained by an optical sensor and by a pressure-based sensor with respect to their ability to represent the true liver motion during breathing.

Methods

A monochrome camera of type IDS uEye (IDS, Obersulm, Germany) equipped with a 4-12 mm vari-focal lens (Tamron, Saitama, Japan) was installed at the rear wall of the MR examination room at approximately 1.9 m from the isocenter of a 3T Ingenia MRI scanner (Philips Healthcare, Best, Netherlands) and used to acquire video data of the subject torso during MR image acquisition (Fig. 1). The camera was operated at 20 frames per second and was connected to a computer via a USB cable equipped with cable traps to reduce RF-interferences. Breathing signals were derived in real-time from the video data using a fully automated algorithm developed previously for patient monitoring3 and streamed to the MR data acquisition system via the scanner physiology interface. This allowed the acquisition of breathing signals simultaneously with both a pressure-based respiratory bellows (Invivo, Gainesville, USA) and the above described optical sensor, that were stored during scanning in physiology log-files for further analysis.

To measure the liver motion due to breathing, dynamic single-slice MR images were acquired in 10 volunteers in sagittal orientation using a single-shot SSFP sequence (pixel size: 1.25x1.25 mm, FOV: 280x280x8 mm, TR / TE = 2.51 / 1.26 ms, flip angle = 20°) with an anterior torso coil-array and a posterior coil-array embedded in the patient support. 120 dynamics with a temporal resolution of 223ms were acquired during regular breathing. The respiratory bellows was positioned on the subject’s abdomen below the anterior torso coil and was secured using a Velcro strap. The study was IRB approved and written informed consent was obtained from all volunteers. During acquisition, timestamps were written in the physiology log-file to mark the acquisition time of the k-space center of each individual dynamic.

Motion fields were computed for each dynamic of the torso scan using the Gunnar Farnebäck optical flow algorithm4 from openCV 2.4.6. The mean displacement along the feet-head direction was computed within a ROI placed manually in the liver, where the motion field computation was found to be the most robust, and used as a marker of the liver motion during breathing [Fig. 2].

Based on the acquisition markers stored in the physiology log-files, breathing signals from the two external sensors were compared to the motion curve derived from the dynamic MR images. Cross-correlation coefficients between the signals were computed to quantify the ability of each sensor to represent liver motion.

Results

In all experiments, signals measured by the external sensors and the liver motion signals computed from the dynamic scans exhibited the expected inspiration and expiration characteristics of breathing motion and showed comparable respiratory patterns, both in amplitude and frequency (Fig. 3). The correlation between liver motion signal and breathing signals measured by the external sensors was higher on average for the camera sensor, with values in the range [0.91-0.98], than for the respiratory bellows (range [0.72-0.94]). In 9 out of 10 cases, the correlation coefficient of the camera signal was larger than that of the respiratory bellows (Fig. 4).

Discussion

Visual inspection of the extracted signals suggested that the higher correlation between camera sensor and liver motion signals was due to a better agreement between peak-to-peak amplitude variations and reduced temporal shift during the expiratory phase. This may indicate that the camera sensor was able to better capture the full torso motion during the expiration, while the respiratory bellows was only sensitive to the local motion of the abdominal wall. Further studies including more subjects with higher variability in morphology and physiology are needed to confirm these preliminary results.

Conclusion

Breathing signals obtained by the optical sensor were found to have high correlation with the computed liver motion and to more accurately represent the breathing status than the pressure-based respiratory bellows in a group of 10 healthy volunteers.

Acknowledgements

No acknowledgement found.

References

  1. J. Maclaren, M. Aksoy, and R. Bammer. Contact-free Physiological Monitoring Using a Markerless Optical System. Magn Reson Med. 2015; 74(2):571-577.
  2. J.Y. Cheng, J. Lu, G. Scott et al. Optical Motion Monitoring for Abdominal and Lung Imaging. ISMRM 2017, #3936.
  3. M. Rocque. Fully Automated Contactless Respiration Monitoring Using a Camera. IEEE International Conference on Consumer Electronics, 2016.
  4. G. Farnebäck. Two-frame motion estimation based on polynomial expansion. In Image Analysis, pages 363-370, Springer, 2003.

Figures

Fig 1: In-bore view from the camera, showing the anterior torso-coil array that was used for the experiments.

Fig 2: Example of computed motion fields overlaid onto the sagittal MR image. The arrows indicate the motion vector at selected pixel positions, with the arrow color coding the displacement amplitude (units are pixels). The ROI used to compute the liver motion statistics is shown as an orange square, and a close-up of the corresponding region is shown in the top-right insert, with axis scale in pixels.

Fig 3: Comparison of the extracted liver motion signal (orange) and the breathing signals measured by the respiratory bellows (purple) and the camera sensor (green) in one subject. The correlation between the liver motion signal curve and each of the sensor signals is indicated in the legend.

Fig. 4: Comparison of cross-correlation coefficients between sensor signal (purple: bellows; green: camera) and liver motion signal for each subject.

Proc. Intl. Soc. Mag. Reson. Med. 26 (2018)
2528