Sep 16 – 20, 2024
Knoxville Convention Center (KCC)
US/Eastern timezone
GRCon24 will take place in Knoxville, TN from Sept 16-20

Motion correction for magnetic resonance imaging using software-defined radio — how the GNU Radio ecosystem facilitates prototyping in academic research

Sep 19, 2024, 1:00 PM
15m
Ballroom AB (Knoxville Convention Center (KCC))

Ballroom AB

Knoxville Convention Center (KCC)

Talk Main Track

Speaker

Dr Christoph Maier (NYU Grossman School of Medicine)

Description

Introduction

Magnetic resonance imaging (MRI) is a key technique in radiology. Its offers excellent soft-tissue contrast and rich insight into the structure and function of the human body, making it an indispensable tool in modern medicine for the diagnosis of many diseases and for treatment monitoring.

The signal amplitude of the underlying nuclear magnetic resonance phenomenon is very weak, however, which results in relatively long acquisition times. This makes the acquisition susceptible to corruption by motion, which is one of the leading causes of non-diagnostic exams in clinical practice. Motion in this context includes both voluntary movement and physiological motion such as the beating heart or respiration.
Imaging can be particularly challenging in the sickest patients, who often struggle with lying still or following breathing instructions.

Various strategies have been developed to handle motion in MRI. These strategies can be divided into prospective techniques, where data is acquired only during a defined motion state, and retrospective techniques, where data is acquired continuously, and motion is compensated for after the scan during the image reconstruction.
Most approaches require a motion signal, which can be derived from the MRI data itself or from external sensors placed in close proximity to the MRI scanner (but operating independently from the imaging apparatus).
Several such motion sensors have been proposed, e.g., ultrasound sensors, optical cameras, radar, and other radiofrequency-based techniques.

In this work, we describe the development of a radar sensor that can be used to monitor human subject motion during an MRI examination, utilizing GNU Radio as a platform for rapid prototyping.

Development and Initial Results

We implemented the radar system using GNU Radio and a networked SDR (USRP N310, Ettus Research).
Two antennas (1x Tx, 1x Rx) were mounted in a bistatic configuration on a half-cylindrical scaffold, which fits over the subject inside the bore of a clinical MRI scanner. We utilized an ultra-wideband antipodal Vivaldi antenna design, which was optimized for near-field microwave imaging at frequencies between 0.6 and 10 GHz. To suppress common-mode currents on the outer shield of the conductor cables induced by the radiofrequency (RF) pulses of the MRI scanner, cable traps were attached to the antenna cables.

A continuous-wave Doppler regime was employed to sense motion, transmitting a 1 kHz tone at various carrier frequencies. The Doppler phase shift between forward and reflected waves served as the motion signal and was derived in real-time using GNU Radio digital signal processing logic.
The radar operated completely independently from the MRI scanner. Therefore, it was required to synchronize the two data streams in time to make use of the radar-derived motion signal for MRI motion correction.
For this purpose, the RF pulses emitted during an imaging scan were captured on the second channel of the SDR tuned to 63.64 MHz, close to the operational frequency of the MRI scanner (i.e., the Larmor frequency of protons at 1.5 T magnetic field strength).

Recorded signal data was archived in the Signal Metadata Format (SigMF), as standardized by the community. This machine-readable format helps in organizing large collections of experimental raw data and facilitates the automated and reproducible generation of scientific results.

Doppler signals were acquired concomitantly with MRI scans in a motion phantom and in human volunteers.
The motion phantom serves as a 'test bed' for the technical development of the radar motion sensor. It was designed and 3D printed in-house and consists of a pendulum moving inside a water tank, simulating internal organ motion. The pendulum is actuated by an Arduino microcontroller connected to a stepper motor (located outside the magnet bore). The pendulum can be readily visualized on MR images in order to correlate its position in space over time with the radar signal.
Using cine MRI of the chest in a human volunteer at a frame rate of 4.4 fps, we tracked the movement of the diaphragm, representing a ground truth for respiratory motion.
The Doppler radar-derived motion signal showed good to excellent agreement with this image-based reference, depending on the carrier frequency and breathing pattern (Pearson's coefficient of correlation: 0.73 - 0.95).
Finally, a high-resolution 3D volumetric MRI scan of the liver was performed with the subject breathing freely. Raw data from this acquisition was retrospectively corrected for respiratory motion based on the radar signal, resulting in improved image sharpness.

Outlook

As proof of concept, we showed that the prototype radar sensor was able to detect and correct for respiratory motion in humans undergoing an MRI scan.
The flexible nature of the GNU Radio platform enabled quick iteration cycles, which has proven very useful in the development and experimental validation of this technique.
Going forward, we plan to explore on-the-fly adaptation and calibration of operational parameters using custom GNU Radio blocks. This would enable tailoring the system to an individual subject, as the propagation and scattering of electromagnetic waves inside the body strongly depend on electromagnetic tissue properties. The influence of individual organs and their motion could thus be more accurately traced.
Further directions for future research are the application of pulsed radar techniques and multiple input/multiple output antenna configurations, as well as as utilization of the GNU radio platform for triggering the MRI scanner based on the internal motion states of the body.

Funding Acknowledgment

Supported through grant funding from the German Research Foundation (DFG, grant no. 512359237) and from the National Institutes of Health (P41 EB017183).

Talk Length 15 Minutes

Primary authors

Dr Christoph Maier (NYU Grossman School of Medicine) Dr George Verghese (NYU Grossman School of Medicine) Dr Eddy Solomon (Weill Cornell Medicine) Prof. Kai Tobias Block (NYU Grossman School of Medicine) Prof. Leeor Alon (NYU Grossman School of Medicine)

Presentation materials

There are no materials yet.