Evaluation of Registration Accuracy for Online MR-Guided Adaptive Planning
R Wu*, Y Ding, P Castillo, M Pham, T Williamsom, L Bronk, A Liu, C Fuller, J Yang, MD Anderson Cancer Center, Houston, TX
Presentations
MO-IePD-TRACK 3-6 (Monday, 7/26/2021) 12:30 PM - 1:00 PM [Eastern Time (GMT-4)]
Purpose: To quantify the accuracy of rigid and deformable image registration (DIR) by using an in-house -developed phantom for online MR-guided adaptive planning with a Unity MR-Linac.
Methods: We developed a phantom to evaluate the registration accuracy between modalities (MR-CT) and within MR (MR-MR). This acrylic phantom consists of a rotational base and 3 inserts of different shapes (circle, oval, and irregular), which were filled with oil, agarose gel, and Styrofoam jelly to simulate fat, prostate, and lung tissues. Deformations and location changes were simulated by changing the inserts. One CT scan and 48 MR scans (T1- weighted short [2-min] and long [6-min], and T2-weighted [2-min and 6-min] scans) were imported into the Monaco treatment planning system for rigid and DIR tests. Contours of all inserts were manually drawn on each image for evaluation. After DIR, deformed contours were compared with manual contours in terms of Dice similarity coefficient (DSC), Hausdorff distance (HD), and mean distance to agreement (MDA).
Results: Rigid registration accuracy was within 2 mm and 2 degree for all scans (CT-MR, MR-MR). For MR-MR, T1 images had better DIR accuracy than T2 images: DSC T1-T1=0.93, HDT1-T1=0.69 cm, and MDA T1-T1=0.12 cm (vs T2: DSC T2-T2=0.86, HDT2-T2=0.12 cm and MDAT2-T2=0.42 cm). For MR-CT, T1 was also more accurate than T2 (T1: DSCT1-CT=0.82, HDT1-CT=1.56 cm and MDAT1-CT=0.32 cm, compared with T2: DSCT2-CT=0.75, HDT2-CT=1.62 cm and MDAT2-CT=0.47 cm). No difference was found in DIR accuracy between the T1 short and long scans.
Conclusion: Different MR sequences can greatly affect DIR accuracy. However, low-quality T1 scans [2-min] can be used without sacrificing DIR accuracy. This phantom allows end-to-end testing for online MR-based adaptive planning.
ePosters
Keywords
Not Applicable / None Entered.
Taxonomy
Not Applicable / None Entered.
Contact Email