Room: Davidson Ballroom A
Purpose: MRI-guided cervical cancer brachytherapy provides unparalleled soft-tissue contrast for target and organ delineation; however, traditional fiducials cannot be used to identify the radiation source path for planning. Instead, a model is manually aligned to the MRI signal void produced by the applicator, providing the source path in-vivo. The purpose of this study is to develop and validate an algorithm to automatically digitize the applicator using MRI, towards the goal of real-time plan optimization.
Methods: Tandem and ring applicators were automatically digitized using anonymized T2-weighted MR images acquired at 1.5 T from 21 brachytherapy fractions including 9 patients. The two-step model matching algorithm was implemented in C++ involving a 2D matched filter to identify the ring center, and a 3D surface model to identify local position. The algorithm requires no manual initialization. Output is a DICOM-RT file containing the source path in-vivo, calculated using the surface model position and a calibration with our treatment planning system. Errors in the algorithm results were calculated as the 3D distances of the tandem tips and ring centers from those identified manually.
Results: Mean execution time of the algorithm was 2.5 s. The algorithm failed for 1 out of 21 images, identifying an air-filled rectum as a ring. For the successful 20 images, meanÂ±SD error of the tandem tip and ring center was 1.2Â±0.7 mm and 1.4Â±1.0 mm, respectively. MeanÂ±SD [x, y, z] signed error components of the tandem tip and ring center were [0.0Â±0.6, 0.4Â±0.5, 0.2Â±1.1] mm and [-0.2Â±0.5, -0.1Â±0.8, 0.7Â±1.3] mm, respectively.
Conclusion: The algorithm shows promise for real-time applicator digitization with mean error <1.5 mm. Biases in the error components suggest that accuracy could be improved by improving the surface model calibration. This algorithm is being refined for an intra-operative planning workflow to be validated in a larger patient cohort.
Brachytherapy, MRI, Image Processing