Functional magnetic resonance imaging (fMRI) is increasingly used for pre-surgical planning, behavioral assessment, and research into neurological and psychiatric illnesses and brain function. However, subject motion and observation noise remain significant impediments to obtaining high-quality fMRI time series data. The proposed research focuses on developing techniques for adjusting the orientation and other parameters of the fMRI acquisition and reconstruction to the noise level of the data and the detected patient motion in real- time.
Our first aim i s twofold: (1) to modify fMRI pulse sequences to automatically reorient the acquired volume to compensate for pre-measured motion and (2) to denoise the acquired data using sparsity-based reconstruction guided by the measured noise level. Initially, we use motion-tracking devices and noise-only pre-scans to measure motion and estimate the noise variance;
our second aim i s to predict motion and estimate the noise level from previous frames in the time series and incorporate these predictions into the adaptive acquisition and reconstruction methods, closing the feedback loop. In addition to developing adaptive versions of echo-planar imaging (EPI) and spiral pulse sequences widely used in fMRI, we propose extending the adaptive approach to accelerated parallel imaging acquisitions to enable higher resolution activation maps. These acquisitions accelerate imaging by undersampling the frequency domain (k-space) and using the redundancy from parallel receiver coils to undo aliasing in the resulting images. Conventional accelerated parallel imaging methods like GRAPPA are particularly susceptible to motion because the calibrated kernels used for interpolating missing k-space frequencies become inaccurate. Accelerated parallel imaging methods also greatly amplify the noise in the data, making statistical analysis unreliable at high accelerations. Thus, the third aim is to adjust the accelerated parallel imaging acquisition and reconstruction for motion and noise amplification to yield greatly improved quality images from substantially undersampled data. Altogether, we aim to provide novel acquisition and reconstruction techniques that are more robust to motion and observation noise and have greater resolution than conventional fMRI.
Functional magnetic resonance imaging (fMRI) has provided neuroscientists and clinicians with an unprecedented new look at brain activity, with applications in neurosurgery, behavioral assessment, and studying brain development and disease. The proposed development of adaptive acquisition and reconstruction techniques will improve the robustness of fMRI time series to subject motion and observation noise, especially for accelerated parallel imaging acquisitions with greater resolution. Such improvements will enable more detailed and reliable functional imaging in populations where subject motion is problematic, such as children.
|Weller, Daniel S; Pnueli, Ayelet; Divon, Gilad et al. (2015) Undersampled Phase Retrieval with Outliers. IEEE Trans Comput Imaging 1:247-258|
|Weller, Daniel S; Ramani, Sathish; Fessler, Jeffrey A (2014) Augmented Lagrangian with variable splitting for faster non-Cartesian L1-SPIRiT MR image reconstruction. IEEE Trans Med Imaging 33:351-61|
|Weller, Daniel S; Ramani, Sathish; Nielsen, Jon-Fredrik et al. (2014) Monte Carlo SURE-based parameter selection for parallel magnetic resonance imaging reconstruction. Magn Reson Med 71:1760-70|
|Ramani, Sathish; Weller, Daniel S; Nielsen, Jon-Fredrik et al. (2013) Non-cartesian MRI reconstruction with automatic regularization Via Monte-Carlo SURE. IEEE Trans Med Imaging 32:1411-22|
|Weller, Daniel S; Polimeni, Jonathan R; Grady, Leo et al. (2013) Sparsity-promoting calibration for GRAPPA accelerated parallel MRI reconstruction. IEEE Trans Med Imaging 32:1325-35|