Interactive Multi-organ Segmentation Based on Multiple Template Deformation
We present a new method for the segmentation of multiple organs (2D or 3D) which enables user inputs for smart contour editing. By extending the work of [1] with user-provided hard constraints that can be optimized globally or locally, we propose an effic
- PDF / 1,497,027 Bytes
- 8 Pages / 439.363 x 666.131 pts Page_size
- 43 Downloads / 147 Views
2 3
Philips Research MediSys, Paris, France Institut Mines-Telecom, Telecom ParisTech, CNRS LTCI, Paris, France H. Mondor Hospital APHP, Medical Imaging Department, Cr´eteil, France 4 A. Trousseau Hospital APHP, Radiology Department, Paris, France
Abstract. We present a new method for the segmentation of multiple organs (2D or 3D) which enables user inputs for smart contour editing. By extending the work of [1] with user-provided hard constraints that can be optimized globally or locally, we propose an efficient and userfriendly solution that ensures consistent feedback to the user interactions. We demonstrate the potential of our approach through a user study with 10 medical imaging experts, aiming at the correction of 4 organ segmentations in 10 CT volumes. We provide quantitative and qualitative analysis of the users’ feedback.
1
Medical Motivation and Overview
Despite constant improvements of fully automatic segmentation approaches, perfect accuracy remains unreachable in many image processing scenarios, especially when inter- and intra-patient variabilities are important. In a clinical context, the possibility of incorporating user corrections is particularly valuable. From a user point of view, the interactions should be: (i) simple, easy to perform, (ii) fast (ideally with real-time feedback), (iii) intuitive (well-behaved algorithm feedback). Designing efficient and user-friendly algorithms meeting these criteria is particularly difficult. Many works on interactive segmentation can be found in the literature. For instance, the live wire technique [2] is a highly interactive approach, close to fully manual 2D delineation. This approach can be extended to 3D and performed in real-time [3], but it remains very time-consuming for the end user. Various methods aim at optimizing globally an energy taking into account image information and user-provided initializations (e.g. through strokes). This problem is often tackled within discrete optimization frameworks [4,5,6,7,8]. The different formulations found in the literature propose different properties in terms of robustness, speed and sensitivity to initialization. Image partitioning from user inputs can also be formulated as a continuous variational problem [9] and include more global and contextual information. Globally optimizing an energy that changes with each new user input can have counter-intuitive effects, as the algorithm forgets previous results while not putting particular emphasis on the c Springer International Publishing Switzerland 2015 N. Navab et al. (Eds.): MICCAI 2015, Part III, LNCS 9351, pp. 55–62, 2015. DOI: 10.1007/978-3-319-24574-4_7
56
R. Gauriau et al. Template models
ϕ1
ϕ2
Multiple template deformation framework Energy with image driven forces and shape regularization
I
Optimized transformations
Non overlapping constraints
User constraints
ose Pose lization initialization
User inputs
Fig. 1. Illustration of the framework principle on a toy example with two objects.
latest inputs. The non-convexity of some segmentation formul
Data Loading...
