Ensemble Learning and Tensor Regularization for Cone-Beam Computed Tomography-Based Pelvic Organ Segmentation

Cone-beam computed tomography (CBCT) is a widely accessible low-dose imaging approach. However, its use in comprehensive anatomy monitoring is hindered by low contrast and artifacts, resulting in difficulty in identifying structure boundaries. In this study, we propose an ensemble deep-learning model to segment post-prostatectomy organs automatically. We utilize the ensemble logic in various modules during the segmentation process to alleviate the impact of low image quality. Specifically, (1) semantic attention was obtained from an ensemble 2.5D You-only-look-once detector to consistently define regions of interest, (2) multiple view-specific two-stream 2.5D segmentation networks were developed, using auxiliary high-quality CT data to aid CBCT segmentation, and (3) a novel tensor-regularized ensemble scheme was proposed to aggregate the estimates from multiple views and regularize the spatial integrity of the final segmentation. A cross validation study achieved Dice similarity coefficient and mean surface distance of 0.779 ± 0.069 and 2.895 ± 1.496 mm for the rectum, and 0.915 ± 0.055 and 1.675 ± 1.311 mm for the bladder, which are comparable to human observer variation.