Low-Dose Mouse Micro-CT Image Segmentation Based on Multi-Resolution Multi-Organ Shape Prior Knowledge Model

Abstract

Automatic segmentation of computed tomography (CT) images of mice is a step toward computer-assisted preclinical image analysis. Due to the low image quality of micro-CT images, fully-automatic methods may not achieve robust segmentation. For this reason, human interventions are needed to achieve higher segmentation accuracy. In this paper, we propose a human interactive segmentation method incorporating anatomical prior knowledge for multiple abdominal organs in mouse micro-CT images. The method automatically fits a multi-organ shape model to the user-sketched partial boundary contours. Segmentation accuracy is validated by comparing the proposed method against existing shape models. The robustness of our proposed method was evaluated with different users. Finally, the results suggest the proposed method generates accurate segmentation with good robustness.