We've Moved!
Visit SDSU’s new digital collections website at https://digitalcollections.sdsu.edu
Description
The purpose of this study was to investigate image differences between deviceless and device-based 4DCT reconstruction. We retrospectively collected data from 35 patients who underwent 4DCT scans for radiation therapy in the thoracic or abdominal region. 4DCT scans were acquired on a GE Lightspeed CT. Two different reconstruction methods were used to reconstruct 4DCT images into 10 phase bins. One used an external breathing monitoring device (RPM, Varian Medical Systems; 4D-RPM) as the breathing surrogate and the other, the deviceless method, used internal anatomy (Smart Deviceless 4D, GE; 4D-DL). We did a blind study with a group of experts who visually inspected all the 4DCTs, paying close attention to motion artifacts in the diaphragm and tumor regions, and quantified image differences in several regions of interest (ROIs), i.e., the diaphragm region, the body, and, for lung cancer patients, a region around the tumor. We also, calculated the fraction of voxels within 20 HU in both 4DCTs in these ROIs and calculated the normalized cross-correlation (Normxcorr), and the structural similarity (SSIM) between deviced based and deviceless-based reconstructed images. In most cases, the result of the blind study was that the images reconstructed with both methods are similar. In the diaphragm region, 46% of cases were judged being not similar. However, neither method was found to be superior. Among all investigated patients and breathing-phase images, the average fraction of voxels within 20 HU was 0.73±0.07 (1 SD) for the body. When the difference range was increased to 50 HU the average fractions was 0.91±0.01. Corresponding numbers for the diaphragm and lung tumor regions of interest (ROIs) were 0.68±0.07 and 0.63±0.16, respectively. The average Normxcorr between images from both reconstructions was 0.97±0.02 for the body, 0.98±0.02 for the diaphragm and 0.97±0.04 for the tumor ROI. Corresponding numbers for the SSIM were 0.97±0.01 for the body, 0.95±0.02 for the diaphragm and 0.94±0.05 for the tumor ROI. In all three ROIs, the calculated image similarity metrics were high. Overall, we found that 4DCT reconstructed by the deviceless and the deviced-based were similar, without a clear trend on when one of them would be superior
