Raphael Prevost

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

Kidney segmentation in 3D CT images allows extracting useful information for nephrologists. For practical use in clinical routine, such an algorithm should be fast, automatic and robust to contrast-agent enhancement and fields of view. By combining and refining state-of-the-art techniques (random forests and template deformation), we demonstrate the possibility of building an algorithm that meets these requirements. Kidneys are localized with random forests following a coarse-to-fine strategy. Their initial positions detected with global contextual information are refined with a cascade of local regression forests. A classification forest is then used to obtain a probabilistic segmentation of both kidneys. The final segmentation is performed with an implicit template deformation algorithm driven by these kidney probability maps. Our method has been validated on a highly heterogeneous database of 233 CT scans from 89 patients. 80% of the kidneys were accurately detected and segmented (Dice coefficient > 0.90) in a few seconds per volume.

Real-Time 3d image segmentation by user-constrained template deformation

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 non-rigid 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.

Kidney detection and real-time segmentation in 3D contrast-enhanced ultrasound images

In this paper, we present an automatic method to segment the kidney in 3D contrast-enhanced ultrasound (CEUS) images. This modality has lately benefited of an increasing interest for diagnosis and intervention planning, as it allows to visualize blood flow in real-time harmlessly for the patient. Our method is composed of two steps: first, the kidney is automatically localized by a novel robust ellipsoid detector; then, segmentation is obtained through the deformation of this ellipsoid with a model-based approach. To cope with low image quality and strong organ variability induced by pathologies, the algorithm allows the user to refine the result by real-time interactions. Our method has been validated on a representative clinical database.

A Generic, Robust and Fully-Automatic Workflow for 3D CT Liver Segmentation

Liver segmentation in 3D CT images is a fundamental step for surgery planning and follow-up. Robustness, automation and speed are required to fulfill this task efficiently. We propose a fully-automatic workflow for liver segmentation built on state-of-the-art algorithmic components to meet these requirements. The liver is first localized using regression forests. A liver probability map is computed, followed by a global-to-local segmentation strategy using a template deformation framework. We evaluate our method on the SLIVER07 reference database and confirm its state-of-the-art results on a large, varied database of 268 CT volumes. This extensive validation demonstrates the robustness of our approach to variable fields of view, liver contrast, shape and pathologies. Our framework is an attractive tradeoff between robustness, accuracy mean distance to ground truth of 1.7mm and computational speedi¾?46s. We also emphasize the genericity and relative simplicity of our framework, which requires very limited liver-specific tuning.

Precise Ultrasound Bone Registration with Learning-Based Segmentation and Speed of Sound Calibration

Ultrasound imaging is increasingly used in navigated surgery and registration-based applications. However, spatial information quality in ultrasound is relatively inferior to other modalities. Main limiting factors for an accurate registration between ultrasound and other modalities are tissue deformation and speed of sound variation throughout the body. The bone surface in ultrasound is a landmark which is less affected by such geometric distortions. In this paper, we present a workflow to accurately register intra-operative ultrasound images to a reference pre-operative CT volume based on an automatic and real-time image processing pipeline. We show that a convolutional neural network is able to produce robust, accurate and fast bone segmentation of such ultrasound images. We also develop a dedicated method to perform online speed of sound calibration by focusing on the bone area and optimizing the appearance of steered compounded images. We provide extensive validation on both phantom and real cadaver data obtaining overall errors under one millimeter.

Patient-specific 3D Ultrasound Simulation Based on Convolutional Ray-tracing and Appearance Optimization

The simulation of medical ultrasound from patient-specific data may improve the planning and execution of interventions e.g. in the field of neurosurgery. However, both the long computation times and the limited realism due to lack of acoustic information from tomographic scans prevent a wide adoption of such a simulation. In this work, we address these problems by proposing a novel efficient ultrasound simulation method based on convolutional ray-tracing which directly takes volumetric image data as input. We show how the required acoustic simulation parameters can be derived from a segmented MRI scan of the patient. We also propose an automatic optimization of ultrasonic simulation parameters and tissue-specific acoustic properties from matching ultrasound and MRI scan data. Both qualitative and quantitative evaluation on a database of 14 neurosurgical patients demonstrate the potential of our approach for clinical use.

Tagged Template Deformation

Model-based approaches are very popular for medical image segmentation as they carry useful prior information on the target structure. Among them, the implicit template deformation framework recently bridged the gap between the efficiency and flexibility of level-set region competition and the robustness of atlas deformation approaches. This paper generalizes this method by introducing the notion of tagged templates. A tagged template is an implicit model in which different subregions are defined. In each of these subregions, specific image features can be used with various confidence levels. The tags can be either set manually or automatically learnt via a process also hereby described. This generalization therefore greatly widens the scope of potential clinical application of implicit template deformation while maintaining its appealing algorithmic efficiency. We show the great potential of our approach in myocardium segmentation of ultrasound images.

Incorporating Shape Variability in Image Segmentation via Implicit Template Deformation

Implicit template deformation is a model-based segmentation framework that was successfully applied in several medical applications. In this paper, we propose a method to learn and use prior knowledge on shape variability in such framework. This shape prior is learnt via an original and dedicated process in which both an optimal template and principal modes of variations are estimated from a collection of shapes. This learning strategy requires neither a pre-alignment of the training shapes nor one-to-one correspondences between shape sample points. We then generalize the implicit template deformation formulation to automatically select the most plausible deformation as a shape prior. This novel framework maintains the two main properties of implicit template deformation: topology preservation and computational efficiency. Our approach can be applied to any organ with a possibly complex shape but fixed topology. We validate our method on myocardium segmentation from cardiac magnetic resonance short-axis images and demonstrate segmentation improvement over standard template deformation.

Kidney Detection and Segmentation in Contrast-Enhanced Ultrasound 3D Images

Contrast-enhanced ultrasound (CEUS) imaging has lately benefited of an increasing interest for diagnosis and intervention planning, as it allows to visualize blood flow in real-time harmlessly for the patient. It complements thus the anatomical information provided by conventional ultrasound (US). This chapter is dedicated to kidney segmentation methods in 3D CEUS images. First we present a generic and fast two-step approach to locate (via a robust ellipsoid estimation algorithm) and segment (using a template deformation framework) the kidney automatically. Then we show how user interactions can be integrated within the algorithm to guide or correct the segmentation in real-time. Finally, we develop a co-segmentation framework that generalizes the aforementioned method and allows the simultaneous use of multiple images (here the CEUS and the US images) to improve the segmentation result. The different approaches are evaluated on a clinical database of 64 volumes.

Fast solver for some computational imaging problems: A regularized weighted least-squares approach

In this paper we propose to solve a range of computational imaging problems under a unified perspective of a regularized weighted least-squares (RWLS) framework. These problems include data smoothing and completion, edge-preserving filtering, gradient-vector flow estimation, and image registration. Although originally very different, they are special cases of the RWLS model using different data weightings and regularization penalties. Numerically, we propose a preconditioned conjugate gradient scheme which is particularly efficient in solving RWLS problems. We provide a detailed analysis of the system conditioning justifying our choice of the preconditioner that improves the convergence. This numerical solver, which is simple, scalable and parallelizable, is found to outperform most of the existing schemes for these imaging problems in terms of convergence rate.