Is Similarity a Useful Tool in Assessing Quality of Prostate Brachytherapy Seed Implants?
D Todor*, M Hagan, M Anscher, Virginia Commonwealth University, Richmond, VAPO-BPC-Exhibit Hall-6 Saturday Room: Exhibit Hall
Purpose: To identify whether geometrical similarity between a pre-plan and post-implant plan is necessary and/or sufficient for a good quality implant?
Methods: A unique cohort of 94 prostate seed implant patients was analyzed. For each patient, an ultrasound (US) pre-implant and two CT post-implant (day 0 and day 30) studies were available. Measures for similarity were created and computed using feature vectors based on two classes of moments: first, invariant to rotation and translation, the second polar-radius moments invariant to rotation, translation and scaling. Both similarity measures were calibrated using controlled perturbations (random and systematic) of seed positions and contours in different size implants thus producing meaningful numerical threshold values used in the clinical analysis.
Results: When similarity is based on moments invariant to translation and rotation only, both seeds and contours exhibit a remarkable lack of similarity when pre- and post-implant plans are compared, with contours faring slightly better than seeds. When scaling invariance is added, resemblance between both implants and contours improved dramatically. When comparing post-implant plans at day 0 and day 30, apart from minor seed migration, the only change in the geometry of the implant should be due to resolution of prostate swelling and should be well describe by scaling, and this is clearly shown by the 83% (78 patients out of 94) similarity. Interestingly there is no correlation between seed similarity and contours similarity, nor does D90 or D90 ratios correlates with seed or contours similarity in any of the comparisons.
Conclusion: In the setting of pre-planed prostate seed implants using pre-loaded needles as delivery method, based on our data we conclude that similarity between pre- and post-implant plan does not correlate with either D90 or analogous similarity metrics for prostate contours, thus questioning the utility of creating 'optimal' preplans.