Fully automatic segmentation of the kidney in CT images: a shape constrained Expectation Maximization approach

Kidney segmentation and volumetric measurement from Computed Tomography (CT) datasets has been proven to be an effective and accurate indicator for renal function in many clinical situations. These include urological treatment decision-making, radiotherapy planning, and estimation of the glomerular filtration rate of living donors [1, 2,3]. CT imaging is widely used for kidney analysis and diagnosis since it provides essential anatomical information, including kidney morphology and renal vessel characteristics (e.g., their number, size, and locations). It also helps in detecting pathologies such as renal lesions
and calcifications.
We have developed a fully automatic kidney segmentation method. The resulting segmentation provides accurate and robust clinical volumetric measurements and may improve diagnosis and preoperative planning with spatial visualization. The automatic segmentation task is challenging due to the overlap of CT intensity values between the kidney and its surrounding organs, e.g, the liver.
Existing methods often require extensive user interaction, assume a fixed prior model, and only apply to individual 2D slices. This results in segmentation variability, inaccurate volumetric measures, and possibly misleading visualizations [