Self-Supervised Adversarial Learning with a Limited Dataset for Electronic Cleansing in Computed Tomographic Colonography: A Preliminary Feasibility Study

SIMPLE SUMMARY: Electronic cleansing (EC) is used for performing a virtual cleansing of the colon on CT colonography (CTC) images for colorectal cancer screening. However, current EC methods have limited accuracy, and traditional deep learning is of limited use in CTC. We evaluated the feasibility of using self-supervised adversarial learning to perform EC on a limited dataset with subvoxel accuracy. A 3D generative adversarial network was pre-trained to perform EC on the CTC datasets of an anthropomorphic colon phantom, and it was fine-tuned to each input case by use of a self-supervised learning scheme. The visually perceived quality of the virtual cleansing by this method compared favorably to that of commercial EC software on the virtual 3D fly-through examinations of 18 clinical CTC cases. Our results indicate that the proposed self-supervised scheme is a potentially effective approach for addressing the remaining technical problems of EC in CTC for colorectal cancer screening. ABSTRACT: Existing electronic cleansing (EC) methods for computed tomographic colonography (CTC) are generally based on image segmentation, which limits their accuracy to that of the underlying voxels. Because of the limitations of the available CTC datasets for training, traditional deep learning is of limited use in EC. The purpose of this study was to evaluate the technical feasibility of using a novel self-supervised adversarial learning scheme to perform EC with a limited training dataset with subvoxel accuracy. A three-dimensional (3D) generative adversarial network (3D GAN) was pre-trained to perform EC on CTC datasets of an anthropomorphic phantom. The 3D GAN was then fine-tuned to each input case by use of the self-supervised scheme. The architecture of the 3D GAN was optimized by use of a phantom study. The visually perceived quality of the virtual cleansing by the resulting 3D GAN compared favorably to that of commercial EC software on the virtual 3D fly-through examinations of 18 clinical CTC cases. Thus, the proposed self-supervised 3D GAN, which can be trained to perform EC on a small dataset without image annotations with subvoxel accuracy, is a potentially effective approach for addressing the remaining technical problems of EC in CTC.