A novel image analysis approach for verifying MLC leaf positions in electronic portal images

A novel image analysis approach for verifying MLC leaf positions in electronic portal images

Proceedings of the 42nd Annual ASTRO Meeting 341 resp. 144% for the planned GTV to median and mean 145% for the Eimulated uncorrected GTV indicatin...

127KB Sizes 1 Downloads 72 Views

Proceedings

of the 42nd Annual ASTRO Meeting

341

resp. 144% for the planned GTV to median and mean 145% for the Eimulated uncorrected GTV indicating dose decrease at the margins. The largest decrease of minimum dose was been for pelvic targets with 68% vs. 47% for targets in the lung, liver (28%) and abdomen (43%). A decrease of minimum dose of more than 27% was observed in 5 of 23 cases (22%). Conclusion: A security margin of 5mm in axial and IOmm in longitudinal direction for PTV-definition cannot prevent substantial decrease of minimum target dose due to target mobility and setup-inaccuracy in general. Evaluation of the whole target volume by CT-simulation should be performed to allow for correction of the stereotactic coordinates and to avoid target miss.

2168

A novel image analysis approach

E. LA.Chancy.

L. Levine. G. S. Tracton,

(lnir~,-vify of North Crrrdiw,

for verifying MLC leaf positions in electronic portal images

L. Potter. S. M. Pizer

Chtrprl Hill,

NC

Purpose: The calibration of MLC leaf positioning systems can drift or suddenly change. The purpose of this work was to investigate image segmentation using a novel deformable MLC model for independent verification of leaf positions in electronic portal images. Materials and Methods: In the general deformable model approach, a shape model of an object of interest is constructed from one or more reference images and the model is used to automatically segment the corresponding object in a target image. In OUTapproach the model is positioned in the framework of a target image and the image intemities are sampled in a way that cuubes the model to be attracted to the corresponding object until the medial axes (middles) of the undeformed model and object are spatially registered. In the general case at this stage the shapes of the model and object will not match. Continued sampling of the image at a tinrr scale causes the model boundary to deform to match the shape of the target object. Differences in shape between the reference object and the target object are determined by comparing the deformed and undeformed models. Our models, called M-reps (medial representations) are constructed from a kit of elementary components called medial atoms. In our model of an MLC, one atom is assigned to each leaf pair in an image of the reference field. Each atom has a midpoint in the middle of a leaf pair with two radial vectors of equal length that point in opposite directions and extend to the edges of the leaf pair. During model deformation in a verification image, atom centers and radial vectors are adjusted to minimize a cost function that includes terms describing the amount of model deformation and the degree of match to the target image. We applied our MLC model to target images acquired with a Siemens BeamView EPID. Selected leaves of a Siemens MLC were moved in I mm increments over a total distance of several cm. Images were acquired after each incremental change to test the sensitivity and reproducibility of our approach. In other experiments multiple leaves were moved in larger increments to test the robustmesa of our method against large changes in field shape. Results: Our method can detect changes in leaf settings as small as I mm with a reproducibility large changes in multiple leaf settings.

of <

I mm and is robust against

Conclusion: The \enhitivity of our approach and the one-to-one correspondence between medial atoms and leaf pairs makes our method well suited for measuring leaf positions and directly localizing mispositioned leaves. Resealrh supported hy NCI KOI CA671 S3 and NC1 PO1 CA47OX2

2169

Effectofb

earn energy in prostate serial tomotherapy

D. Dong, W. Grant. J. McGary, Rtrvlor Co//r,qc, 0f’Mrrlicinr.

B. Teh, E. B. Butler

H0ustor1. TX

Objective: To evaluate the effect of beam energy (6MV, IOMV and l5MV x-rays) in an inverse planning tomotherapy for treatment of prostate cancer patients.

system using serial

Materials and Methods: Seven prostate cancer patients were selected to study the effect of beam energy in Peacock-based tomotherapy delivery. Corvus 3.0 (NOMOS Corp., Sewickley, PA) inverse planning software was used to prescribe a target dose of 7OGy for prostate and seminal vesicles. Dose limits for bladder. colon, pelvis and rectum were 65Gy. 68Gy, 5OGy and 68Gy respectively. All plans used identical partial-volume prescription parameters (minimum. goal. maximum and % volume under goal) for each individual organ. To maintain the same effectiveness of treatment, all plans were normalized so that the mean dose to prostate was kept the same for the same patient when using different beam energies. To emphasize the effect of beam energy same patient plans using 6MV and IOMV were compared with the same patient plan using I5MV. The relative ratios and differences were then averaged for the seven patients selected. Results: l5MV beam is only slightly better in target coverage than 6MV and IOMV beams respectively (0.44% and 0.03% better for prostate and 136% and 0.76% better for seminal vesicles). The cold spot and hot spot in targets are within 2% using different energies for the same patient. There is a slight improvement in target dose uniformity in higher energy beams: the standard deviations are 2.28. 2.08 and 2.08 (Gy) for prostate target and 2.55, 2.48 and 2.34 (Gy) for seminal vesicles target, using 6MV. IOMV and l5MV beams respectively. The volumes above limits for bladder (0.28%), colon (177% or 0,48cc), pelvis (0.58%). and rectum (1X4%) are slightly hetter using hMV, compared with IOMV and l5MV. The mean doses to dose limiting organs are within 1% among different beam energies. The mean dose to non-target tissues is higher in 6MV (7%) and IOMV (I%), compared with l5MV. The beam-on time (Monitor Unit settings) is much higher in 6MV (22%) and in IOMV (6%) than in l5MV because of the natural difference in penetrating power among different beam energies. Plans were evaluated for l-cm delivery mode (better spatial resolution for beam delivery): however, plans with 2.cm mode also showed a similar trend although there were larger variations for individual plans. Conclusions: Treatment plan5 of the same patient using different beam energies exhibit very similar results, which means that tomotherapy treatment technique is insensitive to beam energy selections. There are minor improvements in tumor coverage and dose uniformity in high energy beams, at a slight expense in volumes above limits for critical structure&. Higher energy beam slightly reduces peripheral dose hut greatly reduces beam-on time.