(Tuesday, 7/16/2019) 1:45 PM - 3:45 PM
Room: Stars at Night Ballroom 2-3
Purpose: A 3D deeply supervised attention-aware boosted convolutional neural network (DAB-CNN) is presented as a superior alternative to current state-of-the-art convolutional neural networks (CNN) for semantic CT segmentation.
Methods: Spatial attention gates were incorporated into a novel 3D cascaded CNN framework to prioritize relevant anatomy and suppress redundancies within the network. Attention gates encourage a more parsimonious use of network parameters, but are non-trivial to train. Deep supervision was used to facilitate model convergence, encourage semantically meaningful deep features, and mitigate local minima traps during training. Due to the complexity and size of the network, incremental channel boosting was used to decrease memory usage and mitigate catastrophic forgetting during a continuous learning routine. 120 patients who had definitive prostate radiotherapy were used in this study. Training, validation, and testing followed Kaggle competition rules, with 80 patients used for training, 20 patients used for internal validation, and 20 test patients used to report final accuracies.
Results: The accuracy of each method was assessed based on Dice score compared to manually delineated contours as the gold standard. The accuracy of DAB-CNN was compared to a variation of U-Net, attention-enabled U-Net, boosted U-Net, deeply-supervised U-Net, and deeply supervised attention-enabled U-Net. DAB-CNN achieved mean dice scores of 90.02±1.19, 93.12±1.59, 83.46±1.66, and 72.21±1.68 for the prostate, bladder, rectum, and penile bulb respectively. Comparator p-values indicate that DAB-CNN achieved significantly superior dice scores than all alternative algorithms for the prostate, rectum, and penile bulb.
Conclusion: This study demonstrated that attention-enabled boosted convolutional neural networks using deep supervision are capable of achieving superior prediction accuracy compared to current state-of-the-art automatic segmentation methods.
Keywords
Segmentation, Image Processing, CT
Taxonomy
Contact Email
