Automatic Segmentation of Spleen from Abdominal Computed Tomography Images

I. Gutenko, A. E. Kaufman, M. Barish

CEWIT, 2014

Abstract

Assessment of splenic abnormalities is performed with many imaging modalities, including abdominal computed tomography. Such abnormalities include variations in volume, shape, location and number of organs. Accurate estimation of the size of the spleen is imperative for diagnosing and observing the progression of malignancies, infections, inflammatory conditions, and blood cell diseases. The first step of such assessment is an accurate segmentation of the spleen from the CT scan, which is traditionally performed manually for each slice.  Manual segmentation on the CT scan with large inter-slice distance results in subjective measurements of the volume which is computed from the three longest diameters of the segmented organ. We suggest a method for the automatic segmentation of the spleen from abdominal CT imagery. Our method includes a set of preprocessing steps followed by volumetric graph cut segmentation. Preprocessing steps are crucial in determining an initialization region. Current steps include thresholding of the data, a sequence of binary dilation and erosion stages, and an analysis of the resulting connected components. Based on the approximated location of the spleen in the CT scan, we select a connected component within that region for initialization of the segmentation via the graph cut algorithm. We evaluate performance of our segmentation method, display the resulting rendering with segmentation overlay and manual labelling, and discuss its limitations.  

Credits

M. Barish