Performance benchmarking of liver CT image segmentation and volume estimation Wei Xiong a , Jiayin Zhou b , Qi Tian a , Jimmy J. Liu a , Yingyi Qi c , Wee Kheng Leow c Thazin Han b , Shih-chang Wang b a Institute for Infocomm Research, A*STAR, Singapore, {wxiong, tian, jliu@i2r.a-star.edu.sg} b School of Medicine, Nat’l Univ. of Singapore, Singapore {dnrzjy, dnrth, dnrhead@nus.edu.sg} c School of Computing, Nat’l Univ. of Singapore, Singapore, {qiyingyi, leowwk@comp.nus.edu.sg} ABSTRACT In recent years more and more computer aided diagnosis (CAD) systems are being used routinely in hospitals. Image- based knowledge discovery plays important roles in many CAD applications, which have great potential to be integrated into the next-generation picture archiving and communication systems (PACS). Robust medical image segmentation tools are essentials for such discovery in many CAD applications. In this paper we present a platform with necessary tools for performance benchmarking for algorithms of liver segmentation and volume estimation used for liver transplantation planning. It includes an abdominal computer tomography (CT) image database (DB), annotation tools, a ground truth DB, and performance measure protocols. The proposed architecture is generic and can be used for other organs and imaging modalities. In the current study, approximately 70 sets of abdominal CT images with normal livers have been collected and a user-friendly annotation tool is developed to generate ground truth data for a variety of organs, including 2D contours of liver, two kidneys, spleen, aorta and spinal canal. Abdominal organ segmentation algorithms using 2D atlases and 3D probabilistic atlases can be evaluated on the platform. Preliminary benchmark results from the liver segmentation algorithms which make use of statistical knowledge extracted from the abdominal CT image DB are also reported. We target to increase the CT scans to about 300 sets in the near future and plan to make the DBs built available to medical imaging research community for performance benchmarking of liver segmentation algorithms. Keywords: PACS, Abdominal CT imaging, Liver Segmentation, Performance Benchmark, Knowledge Extraction, Probabilistic Atlas, CAD 1. INTRODUCTION In recent years more and more computer aided diagnosis (CAD) systems are being used routinely in many hospitals and specialist centers. Image-based knowledge discovery plays important roles in many CAD applications, which have great potential to be integrated into the next-generation picture archiving and communication systems (PCAS) 1,2 . In order to extract knowledge from large volumes of medical images stored in PACS systems, medical image databases (DBs) have to be constructed based on needs from specific applications and diseases. During the past years, research efforts have been devoted to build medical image DBs for CAD and content based medical image retrieval (CBMIR), such as the Lung Imaging Database Consortium (LIDC) for lung cancer diagnosis in USA 3 and imageCLEF for CBMIR in Europe. 4,5 Such DBs have been greatly supporting the research community. In general, a typical image-based CAD procedure includes the isolation of region of interest (ROI), feature extraction, pattern analysis and interpretation, and decision-making. Hence, efficient and robust medical image segmentation tools are essential in many CAD applications. With established DBs for specific applications, medical images will be processed and segmented for low-level feature extraction and ground truth data will be prepared. High- level organ knowledge, such as 2D / 3D atlases as well as pathological features, can be computed based on the medical image data and the ground truth built. Once these collections of data and knowledge are available, they can then be used to benchmark various tools and systems for medical image segmentation, classification and retrieval. Research efforts have been put into liver segmentation and volume estimation during the past decade. 6-9 In order to meet clinical needs, accurate and robust abdominal organ segmentation is still very challenging due to 1) complex anatomic layouts and very similar densities for different organs in abdominal region, and 2) large variations in shape and locations of the same organs among humans. Due of the limited amount of liver CT data sets used for each study and the Medical Imaging 2008: PACS and Imaging Informatics, edited by Katherine P. Andriole, Khan M. Siddiqui, Proc. of SPIE Vol. 6919, 69190N, (2008) · 1605-7422/08/$18 · doi: 10.1117/12.770858 Proc. of SPIE Vol. 6919 69190N-1