Catálogo de publicaciones - libros

The recent introduction of variable-velocity control for Continuously Moving Table Magnetic Resonance Imaging has introduced the ability for interactive examination across extended fields of view. A common application of this method is contrast-enhanced angiography, where propagation of an intravenously injected contrast agent is tracked throughout the peripheral vascular tree. Whereas current methods for performing contrast bolus tracking are entirely manual, we discuss the complete automation of this process through the use of real-time image processing. Specifically, we provide a coupled intensity-correction procedure and modified Fast Marching method for rapid segmentation of contrast enhanced vasculature in CE-MRA and discuss the incorporation of this process into a framework for fully automated and adaptive control of table motion for real-time tracking of contrast bolus propagation through the lower peripheral vasculature.

Motion during the acquisition of dynamic contrast enhanced MRI can cause model-fitting errors requiring co-registration. Clinical implementations use a pharmacokinetic model to determine lesion parameters from the contrast passage. The input to the model is the time-intensity plot from a region of interest (ROI) covering the lesion extent. Motion correction meanwhile involves interpolation and smoothing operations thereby affecting the time-intensity plots. This paper explores the trade-offs in applying an elastic matching procedure on the lesion detection and proposes enhancements. The method of choice is the 3D realization of the Demon’s elastic matching procedure. We validate our enhancements using synthesized deformation of stationary datasets that also serve as ground-truth. The framework is tested on 42 human eye datasets. Hence, we show that motion correction is beneficial in improving the model-fit and yet needs enhancements to correct for the intensity reductions during parameter estimation.

The evaluation of fluorescence microscopy images acquired in high-throughput cell phenotype screens constitutes a substantial bottleneck and motivates the development of automated image analysis methods. Here we introduce a computational scheme to process 3D multi-cell time-lapse images as they are produced in large-scale RNAi experiments. We describe an approach to automatically segment, track, and classify cell nuclei into different mitotic phases. This enables automated analysis of the duration of single phases of the cell life cycle and thus the identification of cell cultures that show an abnormal mitotic behavior. Our scheme proves a high accuracy, suggesting a promising future for automating the evaluation of high-throughput experiments.

In medical imaging, finding landmarks that provide biologically meaningful correspondences is often a challenging and time-consuming manual task. In this paper we propose a generic and simple algorithm for landmarking non-cortical brain structures automatically. We use a probabilistic model of the image intensities based on the deformation of a tissue probability map, learned from a training set of hand-landmarked images. In this setting, estimating the location of the landmarks in a new image is equivalent to finding, by likelihood maximization, the ”best” deformation from the tissue probability map to the image. The resulting algorithm is able to handle arbitrary types and numbers of landmarks. We demonstrate our algorithm on the detection of 3 landmarks of the hippocampus in brain MR images.

This paper presents a registration framework based on the polynomial expansion transform. The idea of polynomial expansion is that the image is locally approximated by polynomials at each pixel. Starting with observations of how the coefficients of ideal linear and quadratic polynomials change under translation and affine transformation, algorithms are developed to estimate translation and compute affine and deformable registration between a fixed and a moving image, from the polynomial expansion coefficients. All algorithms can be used for signals of any dimensionality. The algorithms are evaluated on medical data.

The estimation and subsequent use of tissue () parameters at each image location can potentially lead to a more reliable classification of breast tissues. values can be estimated using multiple (typically 3) MRI images of different flip angles. However, breathing and other slight movements can render the highly non-linear estimation procedure error-prone. In this paper, a simultaneous multiple image registration method is proposed to solve this problem. The registration method is built upon the idea of conserving inverse consistency and transitivity among the multiple image transformations. The algorithm is applied to both simulated data and real breast MRI images. The performance is compared with existing pairwise image registration method. The results clearly indicate that the simultaneous multiple image registration algorithm leads to much more accurate estimation.

2D-3D registration for angiographic liver interventions is an unsolved problem mainly because of two reasons. First, a suitable protocol for Computed Tomography Angiography (CTA) to contrast liver arteries is not used in clinical practice. Second, an adequate registration algorithm which addresses the difficult task of aligning deformed vessel structures has not been developed yet. We address the first issue by introducing an CT scanning phase and thus create a strong link between radiologists and interventionalists. The scan visualizes arteries similar to the vasculature captured with an intraoperative C-arm acquiring Digitally Subtracted Angiograms (DSAs). Furthermore, we propose a registration algorithm using the new CT phase that aligns arterial structures in two steps: a) Initialization of one corresponding feature using vessel diameter information, b) optimization on three rotational and one translational parameter to register vessel structures that are represented as centerline graphs. We form a space of good features by iteratively creating new graphs from projected centerline images and by restricting the correspondence search only on branching points (the vertices) of the vessel tree. This algorithm shows good convergence and proves to be robust against deformation changes, which is demonstrated through studies on one phantom and three patients.

2D-3D slice-to-volume registration for abdominal organs like liver is difficult due to the breathing motion and tissue deformation. The purpose of our approach is to ease CT-fluoroscopy (CT-fluoro) based needle insertion for the Radiofrequency Liver Ablation procedure using high resolution contrasted preoperative data. In this case, low signal-to-noise ratio, absence of contrast and additional presence of needle in CT-fluoro makes it difficult to guarantee the solution of any deformable slice-to-volume registration algorithm. In this paper, we first propose a method for creating a set of ground truth (GT) simulation data based on a non-linear deformation of the CT-fluoro volume obtained from real patients. Second, we split the CT-fluoro image and apply intensity based rigid and affine registration to each section. We then propose a novel solution, which consists of intuitive visualization sequences of optimal sub-volumes of preinterventional data based on the registration results. Experiments on synthetic and real patient data and direct feedback of two interventionalists validate our alternative approach.

We propose a new method for CT to fluoroscope registration which is very robust and has a wide capture range. The method relies on the Unscented Kalman Filter to search for an optimal registration solution and on modern commodity graphics cards for fast generation of digitally reconstructed radiographs. We extensively test our method using three different anatomical data sets and compare it with an implementation of the commonly used simplex-based method. The experimental results firmly support that, under the same testing conditions, our proposed technique outperforms the simplex-based method in capture range while providing comparable accuracy and computation time.

3D ultrasound (US) is an emerging new imaging technology that appeals to more and more applications in intraoperative guidance of computer-assisted surgery. In a freehand US imaging system, US probe calibration is typically required to construct a 3D image of the patient’s anatomy from a set of 2D US images. Most of the current calibration techniques concern primarily with the precision and accuracy. However, for computer-assisted surgeries that may require a calibration task inside the operating room (OR), many other important aspects have to be considered besides accuracy. In this paper, we propose a novel system for automated calibration that is optimized for the OR usage with real-time feedback and control of the calibration accuracy. We have also designed a novel N-wire phantom, with greatly reduced complexity to facilitate mass production without compromising the accuracy and robustness.