Spaciousness Filters for Non-contrast CT Volume Segmentation of the Intestine Region for Emergency Ileus Diagnosis

摘要

This paper proposes enhancement filters for shape-specific regions, based on radial structure tensor (RST) analysis, which we name "spaciousness filters". RST analysis can be used in a similar way to Hessian analysis for classifying intensity structures. However, RST is insufficient for enhancing regions having little contrast or non-typical morphology. Our proposed filters enhance such regions by extending the ray search scheme of RST analysis to work as a filter evaluating spaciousness. We show applications to the abdominal CT of ileus patients having specific shapes. The intestines (including small intestines) of those patients consist of air, liquid and feces portions, and are not contrast-enhanced by barium. Enhancement of liquid and walls play key roles in the sufficient segmentation of intestines and division between neighboring regions. Experimental results on 7 clinical cases showed that the proposed intestine segmentation method produced higher Dice score (0.68) than traditional RST analysis (0.44), even without specific refinement processes like machine-learning-based false positive reduction.