Catálogo de publicaciones - libros

In robotic radiosurgery, a photon beam source, moved by a robot arm, is used to ablate tumors. The accuracy of the treatment can be improved by predicting respiratory motion to compensate for system delay. We consider a wavelet-based multiscale autoregressive prediction method. The algorithm is extended by introducing a new exponential averaging parameter and the use of the Moore-Penrose pseudo inverse to cope with long-term signal dependencies and system matrix irregularity, respectively. In test cases, this new algorithm outperforms normalized LMS predictors by as much as 50%. With real patient data, we achieve an improvement of around 5 to 10%.

In this paper, we propose a method based on multiple criteria to assist physicians in planning percutaneous RFA on liver. We explain how we extracted information from literature and interviews with radiologists, and formalized them into geometric constraints. We expose then our method to compute the most suitable needle insertion in two steps: computation of authorized insertion zones and multi-criteria optimization of the trajectory within this zones. We focus on the combination of the criteria to optimize and on the optimization step.

Recently, micro-rotation confocal microscopy has enabled the acquisition of a sequence of slices for a non-adherent living cells where the slices’ positions are roughly controlled by a dielectric-field biological cage. The high resolution volume reconstruction requires then the integration of precise alignment of slice positions. We propose in the Bayesian context, a new method combining both slice positioning and 3D volume reconstruction simultaneously, which leads naturally to an energy minimization procedure of a variational problem. An automatic calibration paradigm via Maximum Likelihood estimation (MLE) principle is used for the relative hyper-parameter determination. We provide finally experimental comparison results on both conventional z-stack confocal images and 3D volume reconstruction from micro-rotation slices of the same non-adherent living cell to show its potential biomedical application.

Automated acquisitions in microscopy may come along with strong illumination artifacts due to poor physical imaging conditions. Such artifacts obviously have direct consequences on the efficiency of an image analysis algorithm and on the quantitative measures. In this paper, we propose a method to correct illumination artifacts on biological images. This correction is based on orthogonal polynomial modeling, combined with stationary maximization criteria. To validate the proposed method we show that we improve particle detection algorithm.

X-ray C-arm fluoroscopy is a natural choice for intra-operative seed localization in prostate brachytherapy. Resolving the correspondence of seeds in the projection images can be modeled as an assignment problem that is NP-hard. Our approach rests on the practical observation that the optimal solution has almost zero cost if the pose of the C-arm is known accurately. This allowed us to to derive an equivalent problem of reduced dimensionality that, with linear programming, can be solved efficiently in polynomial time. Additionally, our method demonstrates significantly increased robustness to C-arm pose errors when compared to the prior art. Because under actual clinical circumstances it is exceedingly difficult to track the C-arm, easing on this constraint has additional practical utility.

3D reconstruction from serial 2D microscopy images depends on non-linear alignment of serial sections. For some structures, such as the neuronal circuitry of the brain, very large images at very high resolution are necessary to permit reconstruction. These very large images prevent the direct use of classical registration methods. We propose in this work a method to deal with the non-linear alignment of arbitrarily large 2D images using the finite support properties of cubic B-splines. After initial affine alignment, each large image is split into a grid of smaller overlapping sub-images, which are individually registered using cubic B-splines transformations. Inside the overlapping regions between neighboring sub-images, the coefficients of the knots controlling the B-splines deformations are blended, to create a virtual large grid of knots for the whole image. The sub-images are resampled individually, using the new coefficients, and assembled together into a final large aligned image. We evaluated the method on a series of large transmission electron microscopy images and our results indicate significant improvements compared to both manual and affine alignment.

Ultramicroscopy, a novel optical tomographic imaging modality related to fluorescence microscopy, allows to acquire cross-sectional slices of small specially prepared biological samples with astounding quality and resolution. However, scattering of the fluorescence light causes the quality to decrease proportional to the depth of the currently imaged plane. Scattering and beam thickness of the excitation laser light cause additional image degradation. We perform a physical simulation of the light scattering in order to define a quantitative function of image quality with respect to depth. This allows us to establish 3D-volumes of quality information in addition to the image data. Volumes are acquired at different orientations of the sample, hence providing complementary regions of high quality. We propose an algorithm for rigid 3D-3D registration of these volumes incorporating voxel quality information, based on maximizing an adapted linear correlation term. The quality ratio of the images is then used, along with the registration result, to create improved volumes of the imaged object. The methods are applied on acquisitions of a mouse brain and mouse embryo to create outstanding three-dimensional reconstructions.

We develop a segmentation technique for dynamic PET incorporating the physiological parameters for different regions via kinetic modeling. We demonstrate the usefulness of our technique on fifteen [C]Raclopride simulated PET images. We show qualitatively and quantitatively that the physiologically based algorithm outperforms two classical segmentation techniques. Further, we derive a formula to compute and visualize the uncertainty encountered during the segmentation.

There is an increasing interest in exploiting the biophysical plausible models to investigate the physiological mechanisms that underlie observed BOLD response. However, most existing studies do not produce reliable model parameter estimates, are not robust due to the linearization of the nonlinear model, and do not perform statistics test to detect functional activation. To overcome these limitations, we developed a general framework for the analysis of fMRI data based on nonlinear physiological models. It performs system dynamics analysis to gain meaningful insight, followed by global sensitivity analysis for model reduction which leads to better system identifiability. Subsequently, a nonlinear filter is used to simultaneously estimate the state and parameter of the dynamic system, and statistics test is performed to derive activation maps based on such model. Furthermore, we investigate the change of the activation maps of these hidden physiological variables with experimental paradigm through time as well.

The thresholded t-map produced by the General Linear Model (GLM) gives an effective summary of activation patterns in functional brain images and is widely used for feature selection in fMRI related classification tasks. As part of a project to build content-based retrieval systems for fMRI images, we have investigated ways to make GLM more adaptive and more robust in dealing with fMRI data from widely differing experiments. In this paper we report on exploration of the Finite Impulse Response model, combined with multiple linear regression, to identify the “locally best Hemodynamic Response Function (HRF) for each voxel” and to simultaneously estimate activation levels corresponding to several stimulus conditions. The goal is to develop a procedure for processing datasets of varying natures. Our experiments show that Finite Impulse Response (FIR) models with a smoothing factor produce better retrieval performance than does the canonical double gamma HRF in terms of retrieval accuracy.