Identification of accurate tumor location and shape is highly important in lung cancer radiotherapy, to improve the treatment quality by reducing dose delivery errors. Because a lung tumor moves with the patient's respiration, breathing motion should be correctly analyzed and predicted during the treatment for prevention of tumor miss or undesirable treatment toxicity. Besides, in Image-Guided Radiation Therapy (IGRT), the tumor motion causes difficulties not only in delivering accurate dose, but also in assuring superior quality of imaging techniques such as four-dimensional (4D) Cone Beam Computed Tomography (CBCT) and 4D Magnetic Resonance Imaging (MRI). Specifically, 4D CBCT used in CBCT IGRT requires precise respiratory signal extraction to avoid burry edges, inaccurate tumor shape, and motion-induced artifacts on the reconstructed CBCT image. 4D MRIs used in MRI-guided radiation therapy typically have low resolution as a tradeoff with field of view, image acquisition time, and image quality. To predict the tumor motion and guarantee the superior quality of the imaging techniques, the dissertation is divided into three parts. The first part describes a new prediction method for respiration-related tumor movements, called Intra- and Inter-fractional variation prediction using Fuzzy Deep Learning (IIFDL). IIFDL clusters the respiratory movements based on breathing similarities, and estimates patients' breathing motion using the proposed predictor, called fuzzy deep learning. The second part of the dissertation includes a novel marker-less binning method for 4D CBCT projections, called Image Registration-based Projection Binning (IRPB), which combines intensity-based feature point detection and trajectory tracking using random sample consensus. IRPB extracts breathing motion and phases by analyzing periodicity of tissue feature point trajectories. The third part the dissertation explains a novel Super-Resolution (SR) method for 4D MRI, called Recurrent Deep Learning-based SR (RDLS), comprised of feature extraction, recurrent nonlinear mapping, and reconstruction. RDLS estimates high-resolution MRIs from low-resolution MRIs according to a specified magnification power.

No description available.

L'imagerie TEP permet de mesurer et visualiser les changements de la distribution biologique des radiopharmaceutiques au sein des organes d'intérêt au court du temps. Ce suivi temporel offre des informations très utiles concernant les processus métaboliques et physiologiques sous-jacents, qui peuvent être extraites grâce à différentes techniques de modélisation cinétique. De plus, un autre avantage de la prise en compte de l'information temporelle dans les acquisitions TEP pour les examens en oncologie thoracique concerne le suivi des mouvements respiratoires. Ces acquisitions permettent de mettre en place des protocoles et des méthodologies visant à corriger leurs effets néfastes à la quantification, et les artefacts associés. L'objectif de ce projet est de développer une méthode de reconstruction permettant de combiner et mettre en oeuvre d'une part les corrections nécessaires à la quantification des données en TEP, et d'autre part la modélisation de la biodistribution du radiotraceur au cours du temps permettant d'obtenir des images paramétriques pour l'oncologie thoracique. Dans un premier temps, une méthodologie de correction des effets de volume partiel intégrant, dans le processus de reconstruction, une déconvolution de Lucy-Richardson associée à un débruitage dans le domaine des ondelettes, a été proposée. Une seconde étude a été consacrée au développement d'une méthodologie combinant une régularisation temporelle des données par l'intermédiaire d'un ensemble de fonctions de base temporelles, avec une méthode de correction des mouvements respiratoires basée sur un modèle élastique. Enfin, dans une troisième étape, le modèle cinétique de Patlak a été intégré dans un algorithme de reconstruction dynamique, et associé à la correction de mouvement afin de permettre la reconstruction directe d'images paramétriques de données thoraciques soumises au mouvement respiratoire. Les paramètres de transformation élastique pour la correction de mouvement ont été calculés à partir des images TEP d'intervalles synchronisés par rapport à l'amplitude de la respiration du patient. Des simulations Monte-Carlo d'un fantôme 4D géométrique avec plusieurs niveaux de statistiques, et du fantôme anthropomorphique NCAT intégrant des courbes d'activités temporelles réalistes pour les différents tissus, ont été réalisées afin de comparer les performances de la méthode de reconstruction paramétrique développée dans ce travail avec une approche 3D standard d'analyse cinétique. L'algorithme proposé a ensuite été testé sur des données cliniques de patients présentant un cancer bronchique non à petites cellules. Enfin, après la validation indépendante de l'algorithme de correction des effets de volume partiel d'une part, et de la reconstruction 4D incorporant la régularisation temporelle d'autre part, sur données simulées et cliniques, ces deux méthodologies ont été associées afin d'optimiser l'estimation de la fonction d'entrée à partir d'une région sanguine des images reconstruites. Les résultats de ce travail démontrent que l'approche de reconstruction paramétrique proposée permet de conserver un niveau de bruit stable dans les régions tumorales lorsque la statistique d'acquisition diminue, contrairement à l'approche d'estimation 3D pour laquelle le niveau de bruit constaté augmente. Ce résultat est intéressant dans l'optique d'une réduction de la durée des intervalles de la reconstruction 4D, permettant ainsi de réduire la durée totale de l'acquisition 4D. De plus, l'utilisation des fonctions d'entrée estimées avec les méthodes de régularisation temporelle proposées ont conduit à améliorer l'estimation des paramètres de Patlak. Enfin, la correction élastique du mouvement amène à une diminution du biais d'estimation des deux paramètres de Patlak, en particulier sur les tumeurs de petites dimensions situées dans des régions sensibles au mouvement respiratoire. / Positron emission tomography (PET) is now considered as the gold standard and the main tool for the diagnosis and therapeutic monitoring of oncology patients, especially due to its quantitative aspects. With the advent of multimodal imaging in combined PET and X-ray CT systems, many methodological developments have been proposed in both pre-processing and data acquisition, image reconstruction, as well as post-processing in order to improve the quantification in PET imaging. Another important aspect of PET imaging is its high temporal resolution and ability to perform dynamic acquisitions, benefiting from the high sensitivity achieved with current systems. PET imaging allows measuring and visualizing changes in the biological distribution of radiopharmaceuticals within the organ of interest over time. This time tracking provides valuable information to physicians on underlying metabolic and physiological processes, which can be extracted using pharmacokinetic modeling. The objective of this project is, by taking advantage of dynamic data in PET/CT imaging, to develop a reconstruction method combining in a single process all the correction methodology required to accurately quantify PET data and, at the same time, include a pharmacokinetic model within the reconstruction in order to create parametric images for applications in oncology. In a first step, a partial volume effect correction methodology integrating, within the reconstruction process, the Lucy-Richardson deconvolution algorithm associated with a wavelet-based denoising method has been introduced. A second study focused on the development of a 4D reconstruction methodology performing temporal regularization of the dataset through a set of temporal basis functions, associated with a respiratory motion correction method based on an elastic deformation model. Finally, in a third step, the Patlak kinetic model has been integrated in a dynamic image reconstruction algorithm and associated with the respiratory motion correction methodology in order to allow the direct reconstruction of parametric images from dynamic thoracic datasets affected by the respiratory motion. The elastic transformation parameters derived for the motion correction have been estimated from respiratory-gated PET images according to the amplitude of the patient respiratory cycle. Monte-carlo simulations of two phantoms, a 4D geometrical phantom, and the anthropomorphic NCAT phantom integrating realistic time activity curves for the different tissues, have been performed in order to compare the performances of the proposed 4D parametric reconstruction algorithm with a standard 3D kinetic analysis approach. The proposed algorithm has then been assessed on clinical datasets of several patients with non small cell lung carcinoma. Finally, following the prior validation of the partial volume effect correction algorithm on one hand, and the 4D reconstruction incorporating the temporal regularization on the other hand, on simulated and clinical datasets, these two methodologies have been associated within the 4D reconstruction algorithm in order to optimize the estimation of image derived input functions. The results of this work show that the proposed direct parametric approach allows to maintain a similar noise level in the tumor regions when the statistic decreases, contrary to the 3D estimation approach for which the observed noise level increases. This result suggests interesting perspectives for the reduction of frame duration reduction of 4D reconstruction, allowing a reduction of the total 4D acquisition duration. In addition, the use of input function estimated with the developed temporal regularization methods led to the improvement of the Patlak parameters estimation. Finally, the elastic respiratory motion correction led to a diminution of the estimation bias of both Patlak parameters, in particular for small lesions located in regions affected by the respiratory motion.

This thesis addresses three problems related to image registration, prediction and tracking, applied to Angiography and Oncology. For image analysis, various probabilistic models have been employed to characterize the image deformations, target motions and state estimations. (i) In Digital Subtraction Angiography (DSA), having a high quality visualization of the blood motion in the vessels is essential both in diagnostic and interventional applications. In order to reduce the inherent movement artifacts in DSA, non-rigid image registration is used before subtracting the mask from the contrast image. DSA image registration is a challenging problem, as it requires non-rigid matching across spatially non-uniform control points, at high speed. We model the problem of sub-pixel matching, as a labeling problem on a non-uniform Markov Random Field (MRF). We use quad-trees in a novel way to generate the non uniform grid structure and optimize the registration cost using graph-cuts technique. The MRF formulation produces a smooth displacement field which results in better artifact reduction than with the conventional approach of independently registering the control points. The above approach is further improved using two models. First, we introduce the concept of pivotal and non-pivotal control points. `Pivotal control points' are nodes in the Markov network that are close to the edges in the mask image, while 'non-pivotal control points' are identified in soft tissue regions. This model leads to a novel MRF framework and energy formulation. Next, we propose a Gaussian MRF model and solve the energy minimization problem for sub-pixel DSA registration using Random Walker (RW). An incremental registration approach is developed using quad-tree based MRF structure and RW, wherein the density of control points is hierarchically increased at each level M depending of the features to be used and the required accuracy. A novel numbering scheme of the control points allows us to reuse the computations done at level M in M + 1. Both the models result in an accelerated performance without compromising on the artifact reduction. We have also provided a CUDA based design of the algorithm, and shown performance acceleration on a GPU. We have tested the approach using 25 clinical data sets, and have presented the results of quantitative analysis and clinical assessment. (ii) In External Beam Radiation Therapy (EBRT), in order to monitor the intra fraction motion of thoracic and abdominal tumors, the lung diaphragm apex can be used as an internal marker. However, tracking the position of the apex from image based observations is a challenging problem, as it undergoes both position and shape variation. We propose a novel approach for tracking the ipsilateral hemidiaphragm apex (IHDA) position on CBCT projection images. We model the diaphragm state as a spatiotemporal MRF, and obtain the trace of the apex by solving an energy minimization problem through graph-cuts. We have tested the approach using 15 clinical data sets and found that this approach outperforms the conventional full search method in terms of accuracy. We have provided a GPU based heterogeneous implementation of the algorithm using CUDA to increase the viability of the approach for clinical use. (iii) In an adaptive radiotherapy system, irrespective of the methods used for target observations there is an inherent latency in the beam control as they involve mechanical movement and processing delays. Hence predicting the target position during `beam on target' is essential to increase the control precision. We propose a novel prediction model (called o set sine model) for the breathing pattern. We use IHDA positions (from CBCT images) as measurements and an Unscented Kalman Filter (UKF) for state estimation. The results based on 15 clinical datasets show that, o set sine model outperforms the state of the art LCM model in terms of prediction accuracy.

Magnetic resonance imaging (MRI), is a noninvasive medical imaging testing techniquewhich is used to produce detailed images of internal structure of the human body, includingbones, muscles, organs, and blood vessels. MRI scanners use large magnets and radiowaves to create images of the body. Cardiac MRI scan helps doctors to detect and monitorcardiac diseases like blood clots, artery blockages, and scar tissue etc. Cardiovasculardisease is a type of disease that affects the heart or the blood vessels.This thesis aims to explore the reconstruction of accelerated cardiovascular MRI datato reconstruct under-sampled MRI data acquired after applying accelerated techniques.The focus of this research is to study and implement deep learning techniques to overcomethe aliasing artifacts caused by accelerated imaging. The results of this study will becompared with fully sampled data acquired with traditional existing techniques such asParallel Imaging (PI) and Compressed Sensing (CS).The primary findings of this study show that the proposed deep learning network caneffectively reconstruct under-sampled cardiovascular MRI data acquired using acceleratedimaging techniques. Many experiments were performed to handle 4D Flow data with limitedmemory for training the network. The network’s performance was found to be comparableto the fully sampled data acquired using traditional imaging techniques such asPI and CS. It is also important to note that this study also aimed to investigate the generalizabilityof the proposed deep learning network, specifically FlowVN, when appliedto different datasets. To explore this aspect, two different models were employed: a pretrainedmodel using previous research data and configurations, and a model trained fromscratch using CMIV data with experiments performed to address limited memory issuesassociated with 4D Flow data.