Full-body CT segmentation using 3D extension of two graph-based methods: a feasibility study

摘要

The paper studies the feasibility of using 3D extensionsudof two state-of-the-art segmentation techniques, the StatisticaludRegion Merging (SRM) method and the EfficientudGraph-based Segmentation (EGS) technique, for automaticudanatomy segmentation on clinical 3D CT images. Theudproposed methods are tested on a dataset of 55 images.udThe test is for segmentation of eight representative tissuesud(lungs, stomach, liver, heart, kidneys, spleen, bones andudthe spinal cord) which are vital for accurate calculation ofudradiation doses. The results are evaluated using the Diceudindex, the Hausdorff distance and the Ht index, a measureudof border error with tolerance t pixels addressing the uncertaintyudin the ground truth. The outcome shows that theud3D-SRM method outperforms 3D-EGS and has a great potentialudto become the method of choice for segmentationudof full-body CT images. Using 3D-SRM, the average Diceudindex, the Hausdorff distance across the 8 tissues, and theudH2 were 0.89, 12.5 mm and 0.93, respectively.