Edge-preserving bilateral filtering for images containing dense objects in CT

In computed tomography (CT), dense objects such as bone, metal implants, and contrast medium, induce cupping and streaking artifacts. To reduce those artifacts, various approaches have a common strategy, which is segmenting datasets into different materials and correcting them separately according to their own physical characteristics. However, in most cases, the severe artifacts hinder the primary segmentation which results in low efficiency of artifact reduction approaches. When taking noise under consideration, the accuracy of segmentation gets even worse. In this work, we applied an edge preserving step based on bilateral filtering between reconstruction and segmentation. A traditional bilateral filter performs noise reduction, and a bilateral edge detector exploits the structural edge information. By incorporating the edge information with noise reduced reconstruction images, a more sophisticated segmentation approach is proposed. Quantitative evaluations of noise reduction and segmentation performance are carried out using simulated and real CT datasets. The results show that our approach can reduce streak artifacts at a primary level, which significantly improves segmentation.