Room: AAPM ePoster Library
Purpose: To provide automatic 6D marker-based IGRT patient setup using planar-images and a 2D to 3D line pattern automatic registration scheme (3DLM).
Methods: Markers on 2D paired-planar-images were identified using an enhanced sobel filter. Marker edges were then skeletonized to draw 2D line patterns on paired-planar-image. Central moment of the marker edges was utilized to reconstruct 2D lines to 3D lines. An interactive closest point (ICP) algorithm was followed to calculate the 6D transformation from the 3D line pattern of the pre-treatment paired-image to the reference 3D line pattern of the pair-DRR. The automatic registration scheme was validated in phantom by comparing the 6D transformation calculated from 3DLM to the 6D transformation from CBCT 3D/3D match with the reference CT. Manual 2D/2D match (MM) transformations [x,y,z] based on pre-treatment kV-pair of prostate patients treated at our institutions were compared with 3DLM transformations. P-value<0.05 of paired t-test is considered significant.
Results: Six phantom measurements were done to compare 3DLM with CBCT 3D/3D match. The mean±SD difference in each dimension was (0.1±1.2mm)(x), (0.5±0.6mm)(y), (0.0±0.8mm)(z), (0.2±0.1°)(Rx), (0.0±0.3°)(Ry), (0.5±0.7°)(Rz), respectively. A total of 21 maker implanted prostate pre-treatment verification paired-KV-images were analyzed. Comparing MM results (translation only) to the re-alignment using the 3DLM (in 6D), the mean±SD translation differences (residual error) were (-0.1±0.4mm,p=0.43)(x), (-0.6±0.4mm,p<0.001)(y), (-0.5±0.5mm,p<0.001)(z), respectively. The mean±SD vector distance from MM shift to 3DLM shift was (1±0.4mm). The rotational misalignments calculated from 3DLM were (-3.9±2.8°)(Rx), (0.8±0.9°)(Ry), (-1.6±2.7°)(Rz), respectively. Paired t-test showed insignificant different in the X-direction. The difference in Y&Z directions were significant.
Conclusion: We demonstrated that kV-pair can be used to obtain 6D transformations for marker-based patient setup. The 3DLM scheme is feasible and has accuracy comparable to the 3D/3D CBCT match. Its clinical implementation is more efficient than taking CBCT and has both translational and rotational misalignment information as using CBCT.
Image-guided Therapy, Edge Detection, Registration