Fast and Robust Semi-automatic Liver Segmentation with Haptic Interaction Erik Vidholm 1 , Sven Nilsson 2 , and Ingela Nystr¨ om 1 1 Centre for Image Analysis, Uppsala University, Sweden erik@cb.uu.se 2 Dept. of Radiology and Clinical Immunology, Uppsala University Hospital, Sweden Abstract. We present a method for semi-automatic segmentation of the liver from CT scans. True 3D interaction with haptic feedback is used to facilitate initialization, i.e., seeding of a fast marching algorithm. Four users initialized 52 datasets and the mean interaction time was 40 seconds. The segmentation accuracy was verified by a radiologist. Volume measurements and segmentation precision show that the method has a high reproducibility. 1 Introduction One of the most important steps in medical image analysis is segmentation, i.e., the process of classifying data elements as object or background. Segmentation is needed in diagnostics, therapy monitoring, surgery planning, and several other medical applications. To manually segment the structures of interest in medical datasets is a very tedious and error-prone procedure, while fully automatic seg- mentation is, despite decades of research, still an unsolved problem. Therefore, many methods are semi-automatic, i.e., the segmentation algorithm is provided with high-level knowledge from the user [1]. A successful semi-automatic method takes advantage of the user’s ability to recognize objects and the ability of the computer to delineate objects. A common recognition task in semi-automatic segmentation is initialization by placement of seed points inside the object of in- terest. The interactive part is highly dependent on the user interface. Interfaces that rely on two-dimensional (2D) interaction have many drawbacks when the data is three-dimensional (3D) since it is not straight-forward how to map 2D interaction into 3D space. It has been shown that by using true 3D interaction with haptic feedback, more efficient semi-automatic methods can be obtained [2]. Liver segmentation is of importance in hepatic surgery planning, where it is a first step in the process of finding vessels and tumours, and the classification of liver segments [3,4]. Liver segmentation may also be useful for monitoring pa- tients with liver metastases, where disease progress is correlated to enlargement of the liver [5, p. 580]. Low contrast between organs and the high shape vari- ability of the liver make automatic segmentation a hard task. In [4], a reference 3D liver model is deformed to fit the liver contour in the image. The method performs well, except for atypical livers where manual interaction is needed. Common for many semi-automatic methods is their slice-based 2 1 2 D nature [6]. R. Larsen, M. Nielsen, and J. Sporring (Eds.): MICCAI 2006, LNCS 4191, pp. 774–781, 2006. c  Springer-Verlag Berlin Heidelberg 2006