(Thursday, 7/16/2020) 10:30 AM - 11:30 AM [Eastern Time (GMT-4)]
Room: Track 3
Purpose: The use of magnetic resonance imaging (MRI) has increased considerably in image-guided high dose-rate (HDR) prostate brachytherapy during the past decades because of its superb soft-tissue contrast for target and normal tissue contouring. Radiation source path reconstruction is a critical procedure in HDR brachytherapy treatment planning. Manually identifying the source path is time consuming. This study is to propose a deep-learning-based method for automatically reconstructing multiple needles for MRI-guided prostate cancer HDR brachytherapy treatment planning.
Methods: We developed a deep learning model using attention gated U-Net for multi-needle segmentation in MRI. The attention gates were used to improve the accuracy of small needle points. Furthermore, total variation (TV) regularization model was incorporated to encode the spatial continuity of needles into the model. The model was trained using paired MR images and their corresponding binary needle annotations (ground truth) provided by experienced physicists. After the network was trained, the locations and sizes of needles in the MR images of a new prostate cancer patient receiving HDR brachytherapy were predicted by the model.
Results: Our method detected 99% of the 167 needles from 11 HDR prostate brachytherapy patients with a needle shaft error of 0.93±0.45 mm and a needle tip error of 0.61±0.98 mm. For tip localization, the proposed method resulted in 80% needles with error less than 1.0 mm, while it achieved 65% localizations with less than 1.0 mm error for shaft localization.
Conclusion: In this study, we proposed a novel multi-needle detection method to precisely localize the tips and shafts of needles in 3D MRI images of HDR prostate brachytherapy. It paves the way for elevating the quality and outcome of MRI-guided HDR prostate brachytherapy.
Keywords
Not Applicable / None Entered.
Taxonomy
Not Applicable / None Entered.
Contact Email
