Real-Time 3D Image Segmentation by User-Constrained Template Deformation

Abstract

We describe an algorithm for 3D interactive image segmentation by non-rigid implicit template deformation, with two main original features. First, our formulation incorporates user input as inside/outside labeled points to drive the deformation and improve both robustness and accuracy. This yields inequality constraints, solved using an Augmented Lagrangian approach. Secondly, a fast implementation of nonrigid template-to-image registration enables interactions with a real-time visual feedback. We validated this generic technique on 21 Contrast-Enhanced Ultrasound images of kidneys and obtained accurate segmentation results (Dice> 0.93) in less than 3 clicks in average.

Automatic Detection and Segmentation of Kidneys in 3D CT Images Using Random Forests

“…Remaining cases were mostly due to pathological kidneys not represented in the training set. Such cases could be quickly corrected by the clinician, since the chosen model-based deformation algorithm [5] allows user interactions. We also emphasize the generality of our framework, that could be as future work extended to other organs.…”

“…We followed the framework introduced in [5] to deform the ellipsoid E. A modelbased approach is here particularly suited because (i) kidneys usually have very smooth shapes, (ii) we want the algorithm to reasonably extrapolate the boundary when the probability map is uncertain. Hereafter we recall the main principles of the adapted model-based deformation algorithm.…”

“…Kidneys' positions are first detected with regression forests following a coarse to fine strategy (Section 2). Then a two-step segmentation is performed on cropped images around the kidneys (Section 3) using (i) a random forest to estimate a probability map of each kidney and (ii) a template deformation algorithm [5] to extract the kidney surface. Experiments and results are detailed in Section 4.…”

Registration of Free-Breathing 3D+t Abdominal Perfusion CT Images via Co-segmentation

“…To do so, we extend the model-based segmentation algorithm proposed by Mory et al [10] (that already proved effective kidney segmentation in CT images [11]) to a co-segmentation method that uses multiples images. This exten-sion was inspired by a paper of Yezzi et al [12], in which they segment a pair of CT/MR images with a single shape.…”

“…Our method is based on implicit template deformation [10], that we extend to a generic tracking method that simultaneously estimates the shape of the kidney and its pose in every frame of the sequence.…”