WE-G-BRD-03: Real-Time Tumor Motion Tracking in Low Field Cine-MR Images

Authors


Abstract

Purpose:

ViewRay's MRIdian system acquires cine-MR images during treatment, allowing for real-time visualization of the tumor without additional dose. We propose a novel method for tracking the tumor target on these cine-MR images in real-time.

Methods:

9 patients with mobile abdominal tumors (3 stomach, 1 pancreas, 1 adrenal, 2 mesenteric, 2 peritoneal) were imaged on ViewRay (0.3T, open-bore; TrueFISP). Sagittal slice cine-MR images (3.5×3.5mm pixels) were acquired at 4 frames per second. A 25-second training period was used to collect 100 images during normal breathing. PCA was used to bin the images using the first 2 eigenvalues. Manual contouring of the target was performed on the first eigenimage. The mean eigenvalues in each bin were used to reconstruct 15 images, on which template matching was used to locate the target. A linear fit of the pixel locations was used to create the motion model. For each subsequently acquired image, the 2-component PCA model was inverted and the values plugged into the motion model to get the corresponding pixel location. Finally, this location and the user-drawn contour were used as the initialization for an active contouring algorithm to refine the edges of the tracked target. 100 images for each patient were manually contoured and used as the gold standard for comparison using the Dice similarity coefficient (DSC) and the modified Hausdorff distance (MHD).

Results:

The mean DSC is 0.87±0.09 and MHD is 2.6±0.9mm, which is less than the pixel size. The algorithm takes 100ms per image using Matlab, and is therefore more than fast enough for real-time use.

Conclusion:

We have developed a robust method for real-time tumor motion tracking in low-resolution, low-field cine-MR images. We foresee our method being used for automated gating and dosimetric measurement purposes with the ViewRay system.

Ancillary