Electronic cleansing (EC) is a method that segments fecal material tagged by an X-ray-opaque oral contrast agent in computed tomographic colonography (CTC) images, and effectively removes the material for digitally cleansing the colon. In this study, we developed a novel EC method, called mosaic decomposition (MD), for reduction of the artifacts due to incomplete cleansing of inhomogeneously tagged fecal material in CTC images, especially in noncathartic CTC images. In our approach, the entire colonic region, including the residual fecal regions, was first decomposed into a set of local homogeneous regions, called tiles, after application of a 3-D watershed transform to the CTC images. Each tile was then subjected to a single-class support vector machine (SVM) classifier for soft-tissue discrimination. The feature set of the soft-tissue SVM classifier was selected by a genetic algorithm (GA). A scalar index, called a soft-tissue likelihood, is formulated for differentiation of the soft-tissue tiles from those of other materials. Then, EC based on MD, called MD-cleansing, is performed by first initializing of the level-set front with the classified tagged regions; the front is then evolved by use of a speed function that was designed, based on the soft-tissue index, to reserve the submerged soft-tissue structures while suppressing the residual fecal regions. The performance of the MD-cleansing method was evaluated by use of a phantom and of clinical cases. In the phantom evaluation, our MD-cleansing was trained with the supine (prone) scan and tested on the prone (supine) scan, respectively. In both cases, the sensitivity and specificity of classification were 100%. The average cleansing ratio was 90.6%, and the soft-tissue preservation ratio was 97.6%. In the clinical evaluation, 10 noncathartic CTC cases (20 scans) were collected, and the ground truth of a total of 2095 tiles was established by manual assignment of a material class to each tile. Five cases were randomly selected for training GA/SVM, and the remaining five cases were used for testing. The overall sensitivity and specificity of the proposed classification scheme were 97.1% and 85.3%, respectively, and the accuracy was 94.6%. The area under the ROC curve (Az) was 0.96. Our results indicated that the use of MD-cleansing substantially improved the effectiveness of our EC method in the reduction of incomplete cleansing artifacts.