Medical Image Analysis (1996/7) volume 1, number 4, pp 331–341 c  Oxford University Press Interactive live-wire boundary extraction William A. Barrett ∗ and Eric N. Mortensen Department of Computer Science, Brigham Young University, Provo, UT 84602, USA Abstract Live-wire segmentation is a new interactive tool for efficient, accurate and reproducible boundary extraction which requires minimal user input with a mouse. Optimal boundaries are computed and selected at interactive rates as the user moves the mouse starting from a manually specified seed point. When the mouse position comes into the proximity of an object edge, a ‘live-wire’ boundary snaps to, and wraps around the object of interest. The input of a new seed point ‘freezes’ the selected boundary segment and the process is repeated until the boundary is complete. Two novel enhancements to the basic live-wire methodology include boundary cooling and on-the-fly training. Data-driven boundary cooling generates seed points automatically and further reduces user input. On-the-fly training adapts the dynamic boundary to edges of current interest. Using the live-wire technique, boundaries are extracted in one-fifth of the time required for manual tracing, but with 4.4 times greater accuracy and 4.8 times greater reproducibility. In particular, interobserver reproducibility using the live-wire tool is 3.8 times greater than intraobserver repro- ducibility using manual tracing. Keywords: boundary cooling and training, speed, accuracy and reproducibility, graph searching, interactive segmentation, live-wire boundary extraction Received November 20, 1996; revised February 6, 1997; accepted February 10, 1997 1. INTRODUCTION Due to the wide variety of image types and content, fully automated segmentation is still an unsolved problem, while manual segmentation is tedious and time consuming, lack- ing in precision and impractical when applied to extensive temporal or spatial sequences of images. However, most current computer-based techniques require significant user input to specify a region of interest, initialize or control the segmentation process or perform subsequent correction to, or adjustment of, boundary points. Thus, to perform image segmentation in a general and practical way, in- telligent, interactive tools must be provided to minimize user input and increase the efficiency and robustness with which accurate, mathematically optimal contours can be extracted. Previous algorithms have incorporated higher- level constraints, but still use local boundary defining cri- teria which increases the susceptibility to noise (O’Brien and Ezquerra, 1994; Gleicher, 1995). Other researchers ∗ Corresponding author (e-mail: barrett@cs.byu.edu) (Montanari, 1971; Chien and Fu, 1974; Martelli, 1976; Bal- lard and Brown, 1982; Pope et al., 1984) have incorpo- rated global properties for robustness and to produce math- ematically optimal boundaries, but most of these methods use one-dimensional (1-D) implementations which impose di- rectional sampling and searching constraints to extract two- dimensional (2-D) boundaries, thus requiring 2-D boundary templates, as with snakes. In addition, many of the former techniques exhibit a high degree of domain dependence and usually still require extensive user interaction to define an initial set of boundary points or region of interest, or to mod- ify results and thereby produce accurate boundaries. Finally, these methods are often very computational or make use of it- erative algorithms which limit high-level human interactivity as well as the speed with which boundaries can be extracted. More recent boundary definition methods make use of ac- tive contours or snakes (Kass et al., 1987; Amini et al., 1990; Williams and Shah, 1992; Daneels et al., 1993), to ‘improve’ a manually entered rough approximation. After being initial- ized with a rough boundary approximation, snakes iteratively adjust boundary points in parallel in an attempt to minimize an energy functional and achieve an optimal boundary. The