Purpose: To aid the organ contouring in high dose rate (HDR) prostate brachytherapy, a deep learning approach was first developed for automatic registration of magnetic resonance (MR) images to treatment planning ultrasound (US) images with severe needle artifacts.
Methods: A new segmentation-based registration framework was proposed for this multi-modality image registration based on a weakly-supervised strategy, to address the problem of lacking training labels and difficulty of accurate registration from inferior image quality. Specifically, two prostate segmentation networks were trained to generate the weak supervision labels from the MR and US images automatically for the registration network training. To improve the registration accuracy, not only the image pair, but also the corresponding prostate probability maps from the segmentation were further fed to the registration network to predict the deformation matrix. A random augmentation method was designed to alleviate the map overfitting problem during the network training. MR data from 184 patient cases and US data from 167 cases were collected from our institution for the MR and US image segmentation networks, and the registration network learning.
Results: The mean DSC, center-of-mass (COM) distance, Hausdorff distance (HD) and averaged symmetric surface distance (ASSD) results for the registration of prostate contours were 0.87±0.05, 1.70±0.89 mm, 7.21±2.07 mm, 1.61±0.64 mm, respectively. By providing the prostate probability map from the segmentation to the registration network, as well as applying the map augmentation method, the evaluation results of the four metrics were all improved, such as an increase of DSC from 0.83±0.08 to 0.86±0.06 and from 0.86±0.06 to 0.87±0.05, respectively.
Conclusion: A novel segmentation-based registration framework was proposed to automatically register prostate MR images to treatment planning US images. It is potential to be applied in current HDR brachytherapy procedures to improve the treatment efficiency and quality.
Funding Support, Disclosures, and Conflict of Interest: This research was supported by a grant from Varian Medical Systems (Palo Alto, CA).