Gregory A. Book MS, Presenter: Nothing to Disclose

Michael Stevens, Abstract Co-Author: Nothing to Disclose

Michal Assaf, Abstract Co-Author: Nothing to Disclose

Godfrey Pearlson, Abstract Co-Author: Nothing to Disclose

Functional MRI and other neuroimaging modalities are susceptible to normal MR artifacts including RF spikes and coil failures, however the most common artifacts are from motion. Real-time motion detection is necessary to allow the scanner operator a chance to repeat a series if motion exceeds a certain limit, and this capability is not available on all MRI scanners. We seek to solve these problems by creating an automated near real-time system to identify and flag potential quality control issues in neuroimaging data. We additionally introduce an algorithm to quantify motion in 3D structural images.

The QC system was built upon the existing Neuroinformatics Database (NiDB) system, and is modular because new QC protocols can be included. Upon receipt of a complete series, the system runs multiple QC checks in parallel by submitting the jobs to a compute cluster, with the results then stored and displayed on a webpage for each imaging study. QC checks include timeseries motion estimation of functional MRI, SNR calculation of timeseries and 3D volumes, and motion detection in 3D structural volumes. fMRI motion estimation and SNR are calculated using FSL. The 3D volume motion detection algorithm calculates the radial average of the FFT of each slice of an image, then takes the average of the linear regression of the resulting power spectra plots, where a more negative R2 value indicates less high frequency signal and therefore more motion.

The NiDB instance in which the QC system was tested stores 43,855 fMRI series and 18,371 3D structural volumes, including the QC results for the respective series. QC results are available 5-10 minutes after the completion of a series. Timeseries motion estimation allowed scan repeats, SNR values indicated coil failures, and R2 value from 3D volumes was consistent with operator identified motion.

Having near real-time QC metrics available allows the MR operator to repeat a scan while the patient is still in the scanner if artifacts are found. Motion detection in 3D images is most useful when collecting multiple structural images within the same scanning session to identify which images should be excluded from further analysis.

Book, G, Stevens, M, Assaf, M, Pearlson, G, Modular Near Real-time Quality Control Analysis for Neuroimaging Data.  Radiological Society of North America 2014 Scientific Assembly and Annual Meeting, - ,Chicago IL. http://archive.rsna.org/2014/14008193.html