Encrypted login | home

Program Information

Implementation of Volumetric Dynamic Keyhole (VDK) Using Normalized Mutual Information Through Respiratory-Phase Matching

no image available
B Lewis

B Lewis*, D Lee , S Kim , T Kim , Virginia Commonwealth University, Richmond, VA

Presentations

TH-EF-605-6 (Thursday, August 3, 2017) 1:00 PM - 3:00 PM Room: 605


Purpose: To introduce a normalized mutual information (nMI) method to the volumetric dynamic keyhole (VDK) data sharing technique for near real-time 4D MR image acquisition.

Methods: The VDK data sharing technique utilizes a library set of volumes matched to respiratory phases to reduce 4D MRI acquisition times for near real-time imaging. To implement this method, the newly acquired, under sampled, volumes must be quickly matched to previously acquired library phases for reconstruction with the data sharing method. This method calculates nMI between the library and under-sampled image volumes (low resolution images) and matches the library phase with the highest scoring nMI. That library phase is then used for the VDK method. The nMI method was quantified using a solid cylinder in 3D space, and calculating the nMI between the original volume and an identical volume, random noise, the original cylinder shifted 5 and 10 pixels to the right, and the original cylinder shifted 5 and 10 pixels out of the page.

Results: nMI values for the cylinder comparison were 0.4208, 0.0001, 0.2883, 0.2025, 0.3455, and 0.2918 for the identical, random, shifted 5 and 10 pixels right, and shifted 5 and 10 pixels into the page. The nMI method correctly matched eight under-sampled image volumes, using as little as 15 of 126 ky and 15 of 135 kz k-space lines, or 1.32% of the total k-space. The calculation took 1.99±0.01 seconds on average.

Conclusion: nMI is an accurate metric for matching respiratory phase of library images and highly under-sampled low-resolution MR volumes. The calculation speed is not yet sufficient for real-time image acquisition, which requires acquisition and reconstruction in less than 0.25 seconds.


Contact Email: