An automatic and robust algorithm for segmentation of three-dimensional medical images

Abstract

Segmentation is a crucial precursor to most medical image analysis applications. This paper presents a new three-dimensional adaptive region growing algorithm for the automatic segmentation of three-dimensional images. The principle of our algorithm is to obtain a satisfactory segment result by self-tuning the homogeneity constraint step by step, which effectively resolves the dilemma of threshold auto-selection. Novel homogeneity and leakage detection criteria are designed to improve accuracy and robustness. Cavities auto-filling algorithm is also proposed to eliminate the interior cavities. Our algorithm was tested by segmenting lungs from 3D throat CT images and compared with manual segmentation and traditional 3D region growing. Results demonstrate that our algorithm greatly outperforms traditional 3D region growing method and its segment result is close to that of manual segmentation.