Impact of non-rigid registration and retrospective image correction (RETROICOR) on detecting BOLD fMRI vasomotor response in the breast
Tess E. Wallace1, Andrew J. Patterson2, Roie Manavaki1, Martin J. Graves1, and Fiona J. Gilbert1

1Department of Radiology, University of Cambridge, Cambridge, United Kingdom, 2Cambridge University Hospitals NHS Foundation Trust, Cambridge, United Kingdom

Synopsis

Physiological fluctuations and motion artifacts are expected to be dominant sources of noise in BOLD fMRI experiments to assess tumor oxygenation and angiogenesis. This work assesses the impact of a non-rigid registration algorithm and retrospective image correction (RETROICOR) on the detection of activation signals in the breast, both at resting state and in response to a modulated respiratory stimulus paradigm. Our results suggest that correction for motion artifacts is associated with a reduction in false-positive activation effects, which can be further improved by the addition of RETROICOR, confirming the importance of these physiological corrections in functional parameter estimation.

Purpose

Physiological fluctuations resulting from cardiac pulsation and respiration are recognized to be a dominant source of noise in blood oxygenation level-dependent (BOLD) fMRI experiments1. There is growing interest in applying BOLD fMRI techniques outside of the brain to assess tumor oxygenation and angiogenesis via vasomotor response to modulated hyperoxic/hypercapnic gas stimuli2,3. Breathing 100% oxygen and carbogen (5% CO2, 95% O2) have opposing effects on vascular tone, as carbon dioxide is a potent vasodilator. However, optical imaging studies have suggested that physiological fluctuations may confound measurement of hemodynamic response4. Furthermore, respiratory motion artifacts are expected to be an additional source of noise, depending on the target site. Physiological noise correction techniques such as RETROspective Image CORrection (RETROICOR) are commonly applied in brain fMRI experiments to improve the statistical significance of activation signals5. The purpose of this work was to investigate the impact of a non-rigid registration algorithm and RETROICOR on the detection of activation signals in the breast, both at resting state (RS) and in response to a modulated respiratory stimulus paradigm.

Methods

Data Acquisition: Functional data was collected from eight healthy female volunteers using a single-shot fast spin echo sequence to acquire dynamic T2-weighted images. Scan parameters were as follows: 3T (MR750, GE Healthcare, Waukesha, WI), TR 4000ms, TE 58ms, BW ±83kHz, matrix size 128x128, FOV 20cm, slice thickness 5mm, single sagittal slice. RS data was acquired as subjects breathed medical air for twelve minutes. The fMRI stimulus design consisted of breathing carbogen interleaved with oxygen in two-minute blocks, for a total of 16 minutes (Figure 1). Physiology was recorded using the scanner’s built in photoplethysmograph and pneumatic belt, and recording was synchronized with the scan acquisition. Data Analysis: Each image series was aligned using a least squares B-spline non-rigid registration algorithm6. Physiological noise components were calculated by fitting a 5th order Fourier series to the image data based on the phase of the cardiac and respiratory cycles relative to the time of each image acquisition. Subsequent analysis was performed on four datasets: uncorrected, RETROICOR-corrected, registered, and registered plus RETROICOR-corrected. A ROI was drawn to eliminate fat in the outer border of the breast and temporal standard deviation was calculated for each pixel in the ROI. Baseline subtraction of the line of best fit through the data was performed to remove linear drift. The first cycle of data from the activated scan was discarded to allow equilibration of the gas inhalation regime. Signal intensity response for each pixel was cross-correlated with a sine and cosine function at the stimulus frequency (0.0042 Hz). All data processing and analysis were performed using Matlab version 8.3 (The Mathworks, Natick, MA). Paired Student’s t-tests were applied to assess the effect of each correction on the mean temporal standard deviation and the difference in median correlation coefficient between RS and activated scans.

Results

A significant reduction in mean temporal standard deviation of RS data was seen for both registration (23%) and RETROICOR (8%) over all subjects (p<0.001). A further 7% reduction in mean temporal standard deviation was seen when RETROICOR was applied to the registered data (p<0.001), shown in Figure 2. A corresponding reduction in median correlation coefficient of RS data was also seen after registration (p=0.06). Activation maps for air-only and oxygen/carbogen scans are shown for a representative volunteer in Figure 3. Overall, there was a significant difference in median correlation coefficient between the RS and activated scans before any correction was applied (p=0.042). However, this difference was significantly improved after registration (p=0.008) and further improved with the addition of RETROICOR (p=0.004), but not with RETROICOR-correction alone (p=0.085), shown in Figure 4.

Discussion

Signal intensity changes induced by the modulated gas stimuli are small, so the temporal signal-to-noise ratio of the time series is critical for detecting activation effects. Respiratory motion was the dominant source of noise in this study, particularly at the border between fat and fibroglandular tissue. Registration of the dynamic series was important in reducing false-positive activation effects, even when motion artifacts were small. Applying RETROICOR after motion correction gave the greatest overall reduction in temporal standard deviation of RS data and the most significant difference between RS and activated scans. Correcting for physiological fluctuations alone was less effective, likely due to movement of fluctuations between pixels, which is not accounted for by RETROICOR. These results demonstrate a reduction of respiratory motion and cardiac pulsation artifacts is associated with changes in activation parameter calculation, confirming the importance of physiological corrections in detecting functional changes in the breast.

Acknowledgements

This work was supported by the NIHR Cambridge Biomedical Research Centre and the Cambridge Experimental Cancer Medicine Centre.

References

1. Kruger G and Glover GH. Physiological noise in oxygen-sensitive Magnetic Resonance Imaging. Magn. Reson. Med. 2001;46:631–37.

2. Rakow-Penner R, Daniel B and Glover GH. Detecting blood oxygen level-dependent (BOLD) contrast in the breast. J. Magn. Reson. Imaging 2010;32:120–29.

3. Neeman M, Dafni H, Bukhari O, et al. In vivo BOLD contrast MRI mapping of subcutaneous vascular function and maturation: validation by intravital microscopy. Magn. Reson. Med. 2001;46:887–98.

4. Carpenter CM, Rakow-Penner R, Jiang S, et al. Monitoring of hemodynamic changes induced in the healthy breast through inspired gas stimuli with MR-guided diffuse optical imaging. Med. Phys. 2010;37:1638–46.

5. Glover GH, Li, TQ and Ress D. Image-based method for retrospective correction of physiological motion effects in fMRI: RETROICOR. Magn. Reson. Med. 2000;44:162–7.

6. Rueckert D, Sonoda LI, Hayes C, et al. Nonrigid registration using free-form deformations: application to breast MR images. IEEE Trans. Med. Imaging 1999;18:712–21.

Figures

Block design for the 16-minute modulated gas fMRI stimulus. The sinusoidal waveform depicts the model used to fit the signal intensity response.

Boxplot showing the effect of registration and RETROICOR on mean temporal standard deviation of RS data.

Activation maps showing magnitude of correlation coefficients (p<0.05) for a representative volunteer, illustrating the effect of each correction on the detection of activation effects for air-only (RS) and oxygen/carbogen data, overlaid on an anatomical image.

Boxplot showing the effect of registration and RETROICOR on median correlation coefficient of air-only (RS) and oxygen/carbogen data.



Proc. Intl. Soc. Mag. Reson. Med. 24 (2016)
4244