Encrypted login | home

Program Information

Automatic Localization and Orientation of Pd-103 Sources in Civasheet Brachytherapy Device

R Farjam

R Farjam*, G Cohen , K Episcopia , A Damato , Memorial Sloan-Kettering Cancer Center, Tenafly, NJ


SU-H4-GePD-T-3 (Sunday, July 30, 2017) 4:30 PM - 5:00 PM Room: Therapy ePoster Lounge

Purpose: To present a methodology for automatic reconstruction of source orientation for the Civasheet which is an array of uniformly mounted unidirectional gold-shielded Pd-103 sources used as a novel implantable device for intraoperative radiotherapy.

Methods: Device reconstruction is performed by source localization followed by source orientation. A thresholding algorithm for source localization is used. Using an optimization process, a lattice structure is then created representing the device curvature where each source is identified by its neighbors. The source orientation is a vector that is perpendicular to all lines connecting the source to its neighbors. CT scans were acquired of phantoms consisting of a 6x6 civasheet placed on plastic holders of various geometries: a flat surface and cylinders with radii of 17, 36 and 65mm. The orientation error (OR = difference between the source orientations found by the algorithm and that measured analytically) was calculated. A simulation study was further conducted to investigate the effect of source mislocalization on orientation estimation.

Results: For flat surface and cylinders with radii of 17mm, 36mm, and 65mm, the average ± one standard deviation of the OR were 1.81°±0.97°, 5.02°±5.02°, 3.10°±2.84°, and 3.52°±3.02°, respectively. For sources located at the corner, ORs were 2.50°±1.14°, 11.68°±0.91°, 5.87°±1.55°, and 4.81°±5.06°, respectively. For edge sources, ORs were 2.31°±051°, 6.56°±4.99°, 4.47°±3.24°, and 4.68°±3.06°, respectively. For the rest, ORs were 1.15°±0.68°, 1.14°±0.47°, 1.03°±0.70°, and 2.24°±0.51°, respectively. The localization algorithm identifies all sources in all cases. We found that one-voxel source mislocalization results in ORs of 5.02°±2.58°, 4.33°±2.36°, and 3.30°±1.53° for sources located at the corner, edge and center, respectively.

Conclusion: Automatic reconstruction allows estimation of source orientation with an average errors ≤ 2.24° for the inner sources, and potentially exceeding 10° for sources at the edges and corners. Further studies are underway to compare these results to manual reconstruction.

Contact Email: