3D Wavelet-based Fusion Techniques for Biomedical Imaging

Hoy en día las técnicas de adquisición de imágenes tridimensionales son comunes en diversas áreas, pero cabe destacar la relevancia que han adquirido en el ámbito de la imagen biomédica, dentro del cual encontramos una amplia gama de técnicas como la microscopía confocal, microscopía de dos fotones, microscopía de fluorescencia mediante lámina de luz, resonancia magnética nuclear, tomografía por emisión de positrones, tomografía de coherencia óptica, ecografía 3D y un largo etcétera. Un denominador común de todas esas aplicaciones es la constante necesidad por aumentar la resolución y la calidad de las imágenes adquiridas. En algunas de dichas técnicas de imagen tridimensional se da una interesante situación: aunque que cada volumen adquirido no contiene información suficiente para representar el objeto bajo estudio dentro de los parámetros de calidad requeridos por algunas aplicaciones finales, el esquema de adquisición permite la obtención de varios volúmenes que representan diferentes vistas de dicho objeto, de tal forma que cada una de las vistas proporciona información complementaria acerca del mismo. En este tipo de situación es posible, mediante la combinación de varias de esas vistas, obtener una mejor comprensión del objeto que a partir de cada una de ellas por separado. En el contexto de esta Tesis Doctoral se ha propuesto, desarrollado y validado una nueva metodología de proceso de imágenes basada en la transformada wavelet disc¬reta para la combinación, o fusión, de varias vistas con información complementaria de un mismo objeto. El método de fusión propuesto aprovecha la capacidad de descom¬posición en escalas y orientaciones de la transformada wavelet discreta para integrar en un solo volumen toda la información distribuida entre el conjunto de vistas adquiridas. El trabajo se centra en dos modalidades diferentes de imagen biomédica que per¬miten obtener tales adquisiciones multi-vista. La primera es una variante de la micro¬scopía de fluorescencia, la microscopía de fluorescencia mediante lámina de luz, que se utiliza para el estudio del desarrollo temprano de embriones vivos en diferentes modelos animales, como el pez cebra o el erizo de mar. La segunda modalidad es la resonancia magnética nuclear con realce tardío, que constituye una valiosa herramienta para evaluar la viabilidad del tejido miocárdico en pacientes con diversas miocardiopatías. Como parte de este trabajo, el método propuesto ha sido aplicado y validado en am¬bas modalidades de imagen. En el caso de la aplicación a microscopía de fluorescencia, los resultados de la fusión muestran un mejor contraste y nivel de detalle en comparación con cualquiera de las vistas individuales y el método no requiere de conocimiento previo acerca la función de dispersión puntual del sistema de imagen. Además, los resultados se han comparado con otros métodos existentes. Con respecto a la aplicación a imagen de resonancia magnética con realce tardío, los volúmenes fusionados resultantes pre-sentan una mejora cuantitativa en la nitidez de las estructuras relevantes y permiten una interpretación más sencilla y completa de la compleja estructura tridimensional del tejido miocárdico en pacientes con cardiopatía isquémica. Para ambas aplicaciones los resultados de esta tesis se encuentran actualmente en uso en los centros clínicos y de investigación con los que el autor ha colaborado durante este trabajo. Además se ha puesto a libre disposición de la comunidad científica la implementación del método de fusión propuesto. Por último, se ha tramitado también una solicitud de patente internacional que cubre el método de visualización desarrollado para la aplicación de Resonancia Magnética Nuclear. Abstract Nowadays three dimensional imaging techniques are common in several fields, but es-pecially in biomedical imaging, where we can find a wide range of techniques including: Laser Scanning Confocal Microscopy, Laser Scanning Two Photon Microscopy, Light Sheet Fluorescence Microscopy, Magnetic Resonance Imaging, Positron Emission To-mography, Optical Coherence Tomography, 3D Ultrasound Imaging, etc. A common denominator of all those applications being the constant need for further increasing resolution and quality of the acquired images. Interestingly, in some of the mentioned three-dimensional imaging techniques a remarkable situation arises: while a single volume does not contain enough information to represent the object being imaged within the quality parameters required by the final application, the acquisition scheme allows recording several volumes which represent different views of a given object, with each of the views providing complementary information. In this kind of situation one can get a better understanding of the object by combining several views instead of looking at each of them separately. Within such context, in this PhD Thesis we propose, develop and test new image processing methodologies based on the discrete wavelet transform for the combination, or fusion, of several views containing complementary information of a given object. The proposed fusion method exploits the scale and orientation decomposition capabil¬ities of the discrete wavelet transform to integrate in a single volume all the available information distributed among the set of acquired views. The work focuses in two different biomedical imaging modalities which provide such multi-view datasets. The first one is a particular fluorescence microscopy technique, Light-Sheet Fluorescence Microscopy, used for imaging and gaining understanding of the early development of live embryos from different animal models (like zebrafish or sea urchin). The second is Delayed Enhancement Magnetic Resonance Imaging, which is a valuable tool for assessing the viability of myocardial tissue on patients suffering from different cardiomyopathies. As part of this work, the proposed method was implemented and then validated on both imaging modalities. For the fluorescence microscopy application, the fusion results show improved contrast and detail discrimination when compared to any of the individual views and the method does not rely on prior knowledge of the system’s point spread function (PSF). Moreover, the results have shown improved performance with respect to previous PSF independent methods. With respect to its application to Delayed Enhancement Magnetic Resonance Imaging, the resulting fused volumes show a quantitative sharpness improvement and enable an easier and more complete interpretation of complex three-dimensional scar and heterogeneous tissue information in ischemic cardiomyopathy patients. In both applications, the results of this thesis are currently in use in the clinical and research centers with which the author collaborated during his work. An imple¬mentation of the fusion method has also been made freely available to the scientific community. Finally, an international patent application has been filed covering the visualization method developed for the Magnetic Resonance Imaging application.

A New Imaging Software To Evaluate The Centration Of Capsulorhexis And Intraocular Lens (IOL) After Cataract Surgery

Purpose: IOL centration and stability after cataract surgery is of high interest for cataract surgeons and IOL-producing companies. We present a new imaging software to evaluate the centration of the rhexis and the centration of the IOL after cataract surgery.Methods: We developed, in collaboration with the Biomedical Imaging Group (BIG), EPFL, Lausanne, a new working tool in order to assess precisely outcomes after IOL-implantation, such as ideal capsulorhexis and IOL-centration. The software is a plug-in of ImageJ, a general-purpose image processing and image-analysis package. The specifications of this software are: evaluation of the rhexis-centration and evaluation the position of the IOL in the posterior chamber. The end points are to analyze the quality of the centration of a rhexis after cataract surgery, the deformation of the rhexis with capsular bag retraction and the centration of the IOL after implantation.Results: This software delivers tools to interactively measure the distances between limbus, IOL and capsulorhexis and its changes over time. The user is invited to adjust nodes of three radial curves for the limbus, rhexis and the optic of the IOL. The radial distances of the curves are computed to evaluate the IOL implantation. The user is also able to define patterns for ideal capsulorhexis and optimal IOL-centration. We are going to present examples of calculations after cataract surgery.Conclusions: Evaluation of the centration of the rhexis and of the IOL after cataract surgery is an important end point for optimal IOL implantation after cataract surgery. Especially multifocal or accommodative lenses need a precise position in the bag with a good stability over time. This software is able to evaluate these parameters just after the surgery but also its changes over time. The results of these evaluations can lead to an optimizing of surgical procedures and materials.

Abnormal Corpus Callosum Integrity in Bipolar Disorder: A Diffusion Tensor Imaging Study

Objective: Abnormalities in the anterior interhemispheric connections provided by the corpus callosum (CC) have long been implicated in bipolar disorder (BID). In this study, we used complementary diffusion tensor imaging methods to study the structural integrity of the CC and localization of potential abnormalities in BD. Methods: Subjects included 33 participants with BID and 40 healthy comparison participants. Fractional anisotropy (FA) measures were compared between groups with region of interest (ROD methods to investigate the anterior, middle, and posterior CC and voxel-based methods to further localize abnormalities. Results: In ROI-based analyses, FA was significantly decreased in the anterior and middle CC in the BID group (p <.05). Voxel-based analyses similarly localized group differences to the genu, rostral body, and anterior midbody of CC (p <.05, corrected). Conclusion: The findings demonstrate abnormalities in the structural integrity of the anterior CC in BID that might contribute to altered interhemispheric connectivity in this disorder.

Intracellular imaging of nanoparticles: is it an elemental mistake to believe what you see?

In order to understand how nanoparticles (NPs <100 nm) interact with cellular systems, potentially causing adverse effects, it is important to be able to detect and localize them within cells. Due to the small size of NPs, transmission electron microscopy (TEM) is an appropriate technique to use for visualizing NPs inside cells, since light microscopy fails to resolve them at a single particle level. However, the presence of other cellular and non-cellular nano-sized structures in TEM cell samples, which may resemble NPs in size, morphology and electron density, can obstruct the precise intracellular identification of NPs. Therefore, elemental analysis is recommended to confirm the presence of NPs inside the cell. The present study highlights the necessity to perform elemental analysis, specifically energy filtering TEM, to confirm intracellular NP localization using the example of quantum dots (QDs). Recently, QDs have gained increased attention due to their fluorescent characteristics, and possible applications for biomedical imaging have been suggested. Nevertheless, potential adverse effects cannot be excluded and some studies point to a correlation between intracellular particle localization and toxic effects. J774.A1 murine macrophage-like cells were exposed to NH2 polyethylene (PEG) QDs and elemental co-localization analysis of two elements present in the QDs (sulfur and cadmium) was performed on putative intracellular QDs with electron spectroscopic imaging (ESI). Both elements were shown on a single particle level and QDs were confirmed to be located inside intracellular vesicles. Nevertheless, ESI analysis showed that not all nano-sized structures, initially identified as QDs, were confirmed. This observation emphasizes the necessity to perform elemental analysis when investigating intracellular NP localization using TEM.

Processing Pipeline for Digitalizing the Lineage Tree of Early Zebrafish Embryogenesis from Multiharmonic Imaging

The reconstruction of the cell lineage tree of early zebrafish embryogenesis requires the use of in-vivo microscopy imaging and image processing strategies. Second (SHG) and third harmonic generation (THG) microscopy observations in unstained zebrafish embryos allows to detect cell divisions and cell membranes from 1-cell to 1K-cell stage. In this article, we present an ad-hoc image processing pipeline for cell tracking and cell membranes segmentation enabling the reconstruction of the early zebrafish cell lineage tree until the 1K-cell stage. This methodology has been used to obtain digital zebrafish embryos allowing to generate a quantitative description of early zebrafish embryogenesis with minute temporal accuracy and μm spatial resolution

Global and local detection of liver steatosis from ultrasound

Liver steatosis is a common disease usually associated with social and genetic factors. Early detection and quantification is important since it can evolve to cirrhosis. Steatosis is usually a diffuse liver disease, since it is globally affected. However, steatosis can also be focal affecting only some foci difficult to discriminate. In both cases, steatosis is detected by laboratorial analysis and visual inspection of ultrasound images of the hepatic parenchyma. Liver biopsy is the most accurate diagnostic method but its invasive nature suggest the use of other non-invasive methods, while visual inspection of the ultrasound images is subjective and prone to error. In this paper a new Computer Aided Diagnosis (CAD) system for steatosis classification and analysis is presented, where the Bayes Factor, obatined from objective intensity and textural features extracted from US images of the liver, is computed in a local or global basis. The main goal is to provide the physician with an application to make it faster and accurate the diagnosis and quantification of steatosis, namely in a screening approach. The results showed an overall accuracy of 93.54% with a sensibility of 95.83% and 85.71% for normal and steatosis class, respectively. The proposed CAD system seemed suitable as a graphical display for steatosis classification and comparison with some of the most recent works in the literature is also presented.

Chronic liver disease staging classification based on ultrasound, clinical and laboratorial data

In this work the identification and diagnosis of various stages of chronic liver disease is addressed. The classification results of a support vector machine, a decision tree and a k-nearest neighbor classifier are compared. Ultrasound image intensity and textural features are jointly used with clinical and laboratorial data in the staging process. The classifiers training is performed by using a population of 97 patients at six different stages of chronic liver disease and a leave-one-out cross-validation strategy. The best results are obtained using the support vector machine with a radial-basis kernel, with 73.20% of overall accuracy. The good performance of the method is a promising indicator that it can be used, in a non invasive way, to provide reliable information about the chronic liver disease staging.

FMRI 3D registration based on Fourier space subsets using neural networks

In this work, we present a neural network (NN) based method designed for 3D rigid-body registration of FMRI time series, which relies on a limited number of Fourier coefficients of the images to be aligned. These coefficients, which are comprised in a small cubic neighborhood located at the first octant of a 3D Fourier space (including the DC component), are then fed into six NN during the learning stage. Each NN yields the estimates of a registration parameter. The proposed method was assessed for 3D rigid-body transformations, using DC neighborhoods of different sizes. The mean absolute registration errors are of approximately 0.030 mm in translations and 0.030 deg in rotations, for the typical motion amplitudes encountered in FMRI studies. The construction of the training set and the learning stage are fast requiring, respectively, 90 s and 1 to 12 s, depending on the number of input and hidden units of the NN. We believe that NN-based approaches to the problem of FMRI registration can be of great interest in the future. For instance, NN relying on limited K-space data (possibly in navigation echoes) can be a valid solution to the problem of prospective (in frame) FMRI registration.

3D-2D image registration by nonlinear regression

We propose a 3D-2D image registration method that relates image features of 2D projection images to the transformation parameters of the 3D image by nonlinear regression. The method is compared with a conventional registration method based on iterative optimization. For evaluation, simulated X-ray images (DRRs) were generated from coronary artery tree models derived from 3D CTA scans. Registration of nine vessel trees was performed, and the alignment quality was measured by the mean target registration error (mTRE). The regression approach was shown to be slightly less accurate, but much more robust than the method based on an iterative optimization approach.

Brain network analysis of patients with Temporal Lobe Epilepsy

Patients with Temporal Lobe Epilepsy (TLE) suffer from widespread subtle white matter abnormalities and abnormal functional connectivity extending beyond the affected lobe, as revealed by Diffusion Tensor MR Imaging, volumetric and functional MRI studies. Diffusion Spectrum Imaging (DSI) is a diffusion imaging technique with high angular resolution for improving the mapping of white matter pathways. In this study, we used DSI, connectivity matrices and topological measures to investigate how the alteration in structural connectivity influences whole brain structural networks. Eleven patients with right-sided TLE and hippocampal sclerosis and 18 controls underwent our DSI protocol at 3T. The cortical and subcortical grey matters were parcellated into 86 regions of interest and the connectivity between every region pair was estimated using global tractography and a connectivity matrix (the adjacency matrix of the structural network). We then compared the networks of patients and controls using topological measures. In patients, we found a higher characteristic path length and a lower clustering coefficient compared to controls. Local measures at node level of the clustering and efficiency showed a significant difference after a multiple comparison correction (Bonferroni). These significant nodes were located within as well outside the temporal lobe, and the localisation of most of them was consistent with regions known to be part of epileptic networks in TLE. Our results show altered connectivity patterns that are concordant with the mapping of functional epileptic networks in patients with TLE. Further studies are needed to establish the relevance of these findings for the propagation of epileptic activity, cognitive deficits in medial TLE and outcome of epilepsy surgery in individual patients.

Effective diagnosis of Alzheimer's disease by means of large margin-based methodology.

BACKGROUND Functional brain images such as Single-Photon Emission Computed Tomography (SPECT) and Positron Emission Tomography (PET) have been widely used to guide the clinicians in the Alzheimer's Disease (AD) diagnosis. However, the subjectivity involved in their evaluation has favoured the development of Computer Aided Diagnosis (CAD) Systems. METHODS It is proposed a novel combination of feature extraction techniques to improve the diagnosis of AD. Firstly, Regions of Interest (ROIs) are selected by means of a t-test carried out on 3D Normalised Mean Square Error (NMSE) features restricted to be located within a predefined brain activation mask. In order to address the small sample-size problem, the dimension of the feature space was further reduced by: Large Margin Nearest Neighbours using a rectangular matrix (LMNN-RECT), Principal Component Analysis (PCA) or Partial Least Squares (PLS) (the two latter also analysed with a LMNN transformation). Regarding the classifiers, kernel Support Vector Machines (SVMs) and LMNN using Euclidean, Mahalanobis and Energy-based metrics were compared. RESULTS Several experiments were conducted in order to evaluate the proposed LMNN-based feature extraction algorithms and its benefits as: i) linear transformation of the PLS or PCA reduced data, ii) feature reduction technique, and iii) classifier (with Euclidean, Mahalanobis or Energy-based methodology). The system was evaluated by means of k-fold cross-validation yielding accuracy, sensitivity and specificity values of 92.78%, 91.07% and 95.12% (for SPECT) and 90.67%, 88% and 93.33% (for PET), respectively, when a NMSE-PLS-LMNN feature extraction method was used in combination with a SVM classifier, thus outperforming recently reported baseline methods. CONCLUSIONS All the proposed methods turned out to be a valid solution for the presented problem. One of the advances is the robustness of the LMNN algorithm that not only provides higher separation rate between the classes but it also makes (in combination with NMSE and PLS) this rate variation more stable. In addition, their generalization ability is another advance since several experiments were performed on two image modalities (SPECT and PET).