Int J CARS (2009) 4:287–297 DOI 10.1007/s11548-009-0293-2 ORIGINAL ARTICLE Liver segmentation by intensity analysis and anatomical information in multi-slice CT images Amir H. Foruzan · Reza Aghaeizadeh Zoroofi · Masatoshi Hori · Yoshinobu Sato Received: 27 February 2008 / Accepted: 1 February 2009 / Published online: 6 March 2009 © CARS 2009 Abstract Purpose Quantitative assessment and essentially segmen- tation of liver and its tumours are of clinical importance in various procedures such as diagnosis, treatment planning, and monitoring. Moreover, segmentation of liver is the basis of further processing such as visualization, liver resection planning, and liver shape analysis. In this paper, we propose an algorithm to estimate an initial liver boundary. Methods The proposed method consists of four steps as fol- lows: first, we compute statistical parameters of liver’s inten- sity range, associated with a large cross-section of liver CT image, utilizing expectation maximization (EM) algorithm. Second, by automatic extraction of ribs and segmentation of the heart, we define a ROI to confine the liver region for the next operations. Third, we propose a double thresholding approach to divide the liver intensity range into two overlap- ping ranges. In this case, based on a decision table, we label an object as a liver candidate or disregard it from the rest of the procedures. Finally, we employ an anatomical based rule to finalize a candidate as a liver tissue. In this case, we propose a color-map transformation scheme to convert the liver gray images into color images. In this way, we attempt to visually differentiate the liver from its surrounding tissues. A. H. Foruzan · R. Aghaeizadeh Zoroofi (B ) Control and Intelligent Processing Center of Excellence, School of Electrical and Computer Engineering, College of Engineering, University of Tehran, Tehran, Iran e-mail: zoroofi@ut.ac.ir M. Hori Department of Radiology, Graduate School of Medicine, Osaka University, Osaka, Japan Y. Sato Division of Image Analysis, Graduate School of Medicine, Osaka University, Osaka, Japan Results We have evaluated the techniques in the presence of 14 randomly selected local datasets as well as all datasets from the MICCAI 2007 Grand Challenge workshop data- base. For the local datasets, the average overlap error and average volume difference were of values of 15.3 and 2.8%, respectively. In the case of the MICCAI datasets, the above values were estimated as 20.3 and -4.5%, respectively. Conclusion The results reveal that the proposed technique is feasible to perform consistent initial liver borders. The boundary might be then employed in an ‘Active Contour’ algorithm to finalize the liver mask. Keywords Liver segmentation · Thresholding · Initial liver border · Liver’s ROI · Heart segmentation Introduction Computer assisted diagnosis/surgery (CAD/CAS) systems are utilized in various clinical applications such as diagnosis, therapy design, and monitoring. In case of hepatic diseases, typical applications include liver resection planning, path planning, thermal ablation design, and tumor grading. Tech- niques such as segmentation, skeletonization, classification, 3D measurement, multiphase registration, and visualization are essential for these CAD/CAS applications. Among the previously mentioned techniques, segmentation is regarded as the main quantification procedure [1, 2]. In spite of the risk of X-ray ionizing radiations, CT is con- sidered as a major imaging modality for liver diagnostic and treatment. A radiologist employs a CT dataset with a reso- lution of 1 × 1 × 3 mm 3 or less to detect vessels and lesions in liver, leading to approximately 150–300 slices per case. Manual segmentation of the resultant dataset by an expert operator is a time consuming and non-repeatable task [3]. 123