Iterative reconstruction in CT : using mathematical model observers to determine low dose images' trustworthiness

La tomodensitométrie (TDM) est une technique d'imagerie pour laquelle l'intérêt n'a cessé de croitre depuis son apparition au début des années 70. De nos jours, l'utilisation de cette technique est devenue incontournable, grâce entre autres à sa capacité à produire des images diagnostiques de haute qualité. Toutefois, et en dépit d'un bénéfice indiscutable sur la prise en charge des patients, l'augmentation importante du nombre d'examens TDM pratiqués soulève des questions sur l'effet potentiellement dangereux des rayonnements ionisants sur la population. Parmi ces effets néfastes, l'induction de cancers liés à l'exposition aux rayonnements ionisants reste l'un des risques majeurs. Afin que le rapport bénéfice-risques reste favorable au patient il est donc nécessaire de s'assurer que la dose délivrée permette de formuler le bon diagnostic tout en évitant d'avoir recours à des images dont la qualité est inutilement élevée. Ce processus d'optimisation, qui est une préoccupation importante pour les patients adultes, doit même devenir une priorité lorsque l'on examine des enfants ou des adolescents, en particulier lors d'études de suivi requérant plusieurs examens tout au long de leur vie. Enfants et jeunes adultes sont en effet beaucoup plus sensibles aux radiations du fait de leur métabolisme plus rapide que celui des adultes. De plus, les probabilités des évènements auxquels ils s'exposent sont également plus grandes du fait de leur plus longue espérance de vie. L'introduction des algorithmes de reconstruction itératifs, conçus pour réduire l'exposition des patients, est certainement l'une des plus grandes avancées en TDM, mais elle s'accompagne de certaines difficultés en ce qui concerne l'évaluation de la qualité des images produites. Le but de ce travail est de mettre en place une stratégie pour investiguer le potentiel des algorithmes itératifs vis-à-vis de la réduction de dose sans pour autant compromettre la qualité du diagnostic. La difficulté de cette tâche réside principalement dans le fait de disposer d'une méthode visant à évaluer la qualité d'image de façon pertinente d'un point de vue clinique. La première étape a consisté à caractériser la qualité d'image lors d'examen musculo-squelettique. Ce travail a été réalisé en étroite collaboration avec des radiologues pour s'assurer un choix pertinent de critères de qualité d'image. Une attention particulière a été portée au bruit et à la résolution des images reconstruites à l'aide d'algorithmes itératifs. L'analyse de ces paramètres a permis aux radiologues d'adapter leurs protocoles grâce à une possible estimation de la perte de qualité d'image liée à la réduction de dose. Notre travail nous a également permis d'investiguer la diminution de la détectabilité à bas contraste associée à une diminution de la dose ; difficulté majeure lorsque l'on pratique un examen dans la région abdominale. Sachant que des alternatives à la façon standard de caractériser la qualité d'image (métriques de l'espace Fourier) devaient être utilisées, nous nous sommes appuyés sur l'utilisation de modèles d'observateurs mathématiques. Nos paramètres expérimentaux ont ensuite permis de déterminer le type de modèle à utiliser. Les modèles idéaux ont été utilisés pour caractériser la qualité d'image lorsque des paramètres purement physiques concernant la détectabilité du signal devaient être estimés alors que les modèles anthropomorphes ont été utilisés dans des contextes cliniques où les résultats devaient être comparés à ceux d'observateurs humain, tirant profit des propriétés de ce type de modèles. Cette étude a confirmé que l'utilisation de modèles d'observateurs permettait d'évaluer la qualité d'image en utilisant une approche basée sur la tâche à effectuer, permettant ainsi d'établir un lien entre les physiciens médicaux et les radiologues. Nous avons également montré que les reconstructions itératives ont le potentiel de réduire la dose sans altérer la qualité du diagnostic. Parmi les différentes reconstructions itératives, celles de type « model-based » sont celles qui offrent le plus grand potentiel d'optimisation, puisque les images produites grâce à cette modalité conduisent à un diagnostic exact même lors d'acquisitions à très basse dose. Ce travail a également permis de clarifier le rôle du physicien médical en TDM: Les métriques standards restent utiles pour évaluer la conformité d'un appareil aux requis légaux, mais l'utilisation de modèles d'observateurs est inévitable pour optimiser les protocoles d'imagerie. -- Computed tomography (CT) is an imaging technique in which interest has been quickly growing since it began to be used in the 1970s. Today, it has become an extensively used modality because of its ability to produce accurate diagnostic images. However, even if a direct benefit to patient healthcare is attributed to CT, the dramatic increase in the number of CT examinations performed has raised concerns about the potential negative effects of ionising radiation on the population. Among those negative effects, one of the major risks remaining is the development of cancers associated with exposure to diagnostic X-ray procedures. In order to ensure that the benefits-risk ratio still remains in favour of the patient, it is necessary to make sure that the delivered dose leads to the proper diagnosis without producing unnecessarily high-quality images. This optimisation scheme is already an important concern for adult patients, but it must become an even greater priority when examinations are performed on children or young adults, in particular with follow-up studies which require several CT procedures over the patient's life. Indeed, children and young adults are more sensitive to radiation due to their faster metabolism. In addition, harmful consequences have a higher probability to occur because of a younger patient's longer life expectancy. The recent introduction of iterative reconstruction algorithms, which were designed to substantially reduce dose, is certainly a major achievement in CT evolution, but it has also created difficulties in the quality assessment of the images produced using those algorithms. The goal of the present work was to propose a strategy to investigate the potential of iterative reconstructions to reduce dose without compromising the ability to answer the diagnostic questions. The major difficulty entails disposing a clinically relevant way to estimate image quality. To ensure the choice of pertinent image quality criteria this work was continuously performed in close collaboration with radiologists. The work began by tackling the way to characterise image quality when dealing with musculo-skeletal examinations. We focused, in particular, on image noise and spatial resolution behaviours when iterative image reconstruction was used. The analyses of the physical parameters allowed radiologists to adapt their image acquisition and reconstruction protocols while knowing what loss of image quality to expect. This work also dealt with the loss of low-contrast detectability associated with dose reduction, something which is a major concern when dealing with patient dose reduction in abdominal investigations. Knowing that alternative ways had to be used to assess image quality rather than classical Fourier-space metrics, we focused on the use of mathematical model observers. Our experimental parameters determined the type of model to use. Ideal model observers were applied to characterise image quality when purely objective results about the signal detectability were researched, whereas anthropomorphic model observers were used in a more clinical context, when the results had to be compared with the eye of a radiologist thus taking advantage of their incorporation of human visual system elements. This work confirmed that the use of model observers makes it possible to assess image quality using a task-based approach, which, in turn, establishes a bridge between medical physicists and radiologists. It also demonstrated that statistical iterative reconstructions have the potential to reduce the delivered dose without impairing the quality of the diagnosis. Among the different types of iterative reconstructions, model-based ones offer the greatest potential, since images produced using this modality can still lead to an accurate diagnosis even when acquired at very low dose. This work has clarified the role of medical physicists when dealing with CT imaging. The use of the standard metrics used in the field of CT imaging remains quite important when dealing with the assessment of unit compliance to legal requirements, but the use of a model observer is the way to go when dealing with the optimisation of the imaging protocols.

Bayesian Estimation of Noise

In mathematical modeling the estimation of the model parameters is one of the most common problems. The goal is to seek parameters that fit to the measurements as well as possible. There is always error in the measurements which implies uncertainty to the model estimates. In Bayesian statistics all the unknown quantities are presented as probability distributions. If there is knowledge about parameters beforehand, it can be formulated as a prior distribution. The Bays’ rule combines the prior and the measurements to posterior distribution. Mathematical models are typically nonlinear, to produce statistics for them requires efficient sampling algorithms. In this thesis both Metropolis-Hastings (MH), Adaptive Metropolis (AM) algorithms and Gibbs sampling are introduced. In the thesis different ways to present prior distributions are introduced. The main issue is in the measurement error estimation and how to obtain prior knowledge for variance or covariance. Variance and covariance sampling is combined with the algorithms above. The examples of the hyperprior models are applied to estimation of model parameters and error in an outlier case.

A correlation sensitivity analysis of non-life underwriting risk in solvency capital requirement estimation

This paper analyses the impact of using different correlation assumptions between lines of business when estimating the risk-based capital reserve, the Solvency Capital Requirement -SCR-, under Solvency II regulations. A case study is presented and the SCR is calculated according to the Standard Model approach. Alternatively, the requirement is then calculated using an Internal Model based on a Monte Carlo simulation of the net underwriting result at a one-year horizon, with copulas being used to model the dependence between lines of business. To address the impact of these model assumptions on the SCR we conduct a sensitivity analysis. We examine changes in the correlation matrix between lines of business and address the choice of copulas. Drawing on aggregate historical data from the Spanish non-life insurance market between 2000 and 2009, we conclude that modifications of the correlation and dependence assumptions have a significant impact on SCR estimation.

Statistical Analysis of an SEIR Epidemic Model

This thesis was focussed on statistical analysis methods and proposes the use of Bayesian inference to extract information contained in experimental data by estimating Ebola model parameters. The model is a system of differential equations expressing the behavior and dynamics of Ebola. Two sets of data (onset and death data) were both used to estimate parameters, which has not been done by previous researchers in (Chowell, 2004). To be able to use both data, a new version of the model has been built. Model parameters have been estimated and then used to calculate the basic reproduction number and to study the disease-free equilibrium. Estimates of the parameters were useful to determine how well the model fits the data and how good estimates were, in terms of the information they provided about the possible relationship between variables. The solution showed that Ebola model fits the observed onset data at 98.95% and the observed death data at 93.6%. Since Bayesian inference can not be performed analytically, the Markov chain Monte Carlo approach has been used to generate samples from the posterior distribution over parameters. Samples have been used to check the accuracy of the model and other characteristics of the target posteriors.

Models for the estimation of thermodynamic properties of layered double hydroxides: application to the study of their anion exchange characteristics

Several models for the estimation of thermodynamic properties of layered double hydroxides (LDHs) are presented. The predicted thermodynamic quantities calculated by the proposed models agree with experimental thermodynamic data. A thermodynamic study of the anion exchange process on LDHs is also made using the described models. Tables for the prediction of monovalent anion exchange selectivities on LDHs are provided. Reasonable agreement is found between the predicted and the experimental monovalent anion exchange selectivities.

Estimation of age- and stage-specific Catalan breast cancer survival functions using US and Catalan survival data

During the last part of the 1990s the chance of surviving breast cancer increased. Changes in survival functions reflect a mixture of effects. Both, the introduction of adjuvant treatments and early screening with mammography played a role in the decline in mortality. Evaluating the contribution of these interventions using mathematical models requires survival functions before and after their introduction. Furthermore, required survival functions may be different by age groups and are related to disease stage at diagnosis. Sometimes detailed information is not available, as was the case for the region of Catalonia (Spain). Then one may derive the functions using information from other geographical areas. This work presents the methodology used to estimate age- and stage-specific Catalan breast cancer survival functions from scarce Catalan survival data by adapting the age- and stage-specific US functions. Methods: Cubic splines were used to smooth data and obtain continuous hazard rate functions. After, we fitted a Poisson model to derive hazard ratios. The model included time as a covariate. Then the hazard ratios were applied to US survival functions detailed by age and stage to obtain Catalan estimations. Results: We started estimating the hazard ratios for Catalonia versus the USA before and after the introduction of screening. The hazard ratios were then multiplied by the age- and stage-specific breast cancer hazard rates from the USA to obtain the Catalan hazard rates. We also compared breast cancer survival in Catalonia and the USA in two time periods, before cancer control interventions (USA 1975–79, Catalonia 1980–89) and after (USA and Catalonia 1990–2001). Survival in Catalonia in the 1980–89 period was worse than in the USA during 1975–79, but the differences disappeared in 1990–2001. Conclusion: Our results suggest that access to better treatments and quality of care contributed to large improvements in survival in Catalonia. On the other hand, we obtained detailed breast cancer survival functions that will be used for modeling the effect of screening and adjuvant treatments in Catalonia

MCMC Analysis Of Classical Time Series Algorithms

Identification of order of an Autoregressive Moving Average Model (ARMA) by the usual graphical method is subjective. Hence, there is a need of developing a technique to identify the order without employing the graphical investigation of series autocorrelations. To avoid subjectivity, this thesis focuses on determining the order of the Autoregressive Moving Average Model using Reversible Jump Markov Chain Monte Carlo (RJMCMC). The RJMCMC selects the model from a set of the models suggested by better fitting, standard deviation errors and the frequency of accepted data. Together with deep analysis of the classical Box-Jenkins modeling methodology the integration with MCMC algorithms has been focused through parameter estimation and model fitting of ARMA models. This helps to verify how well the MCMC algorithms can treat the ARMA models, by comparing the results with graphical method. It has been seen that the MCMC produced better results than the classical time series approach.

Indomethacin solubility estimation in 1,4-dioxane + water mixtures by the extended hildebrand solubility approach

Extended Hildebrand Solubility Approach (EHSA) was successfully applied to evaluate the solubility of Indomethacin in 1,4-dioxane + water mixtures at 298.15 K. An acceptable correlation-performance of EHSA was found by using a regular polynomial model in order four of the W interaction parameter vs. solubility parameter of the mixtures (overall deviation was 8.9%). Although the mean deviation obtained was similar to that obtained directly by means of an empiric regression of the experimental solubility vs. mixtures solubility parameters, the advantages of EHSA are evident because it requires physicochemical properties easily available for drugs.

Kinetic modeling of mechanisms of industrially important organic reactions in gas and liquid phase

This dissertation is based on 5 articles which deal with reaction mechanisms of the following selected industrially important organic reactions: 1. dehydrocyclization of n-butylbenzene to produce naphthalene 2. dehydrocyclization of 1-(p-tolyl)-2-methylbutane (MB) to produce 2,6-dimethylnaphthalene 3. esterification of neopentyl glycol (NPG) with different carboxylic acids to produce monoesters 4. skeletal isomerization of 1-pentene to produce 2-methyl-1-butene and 2-methyl-2-butene The results of initial- and integral-rate experiments of n-butylbenzene dehydrocyclization over selfmade chromia/alumina catalyst were applied when investigating reaction 2. Reaction 2 was performed using commercial chromia/alumina of different acidity, platina on silica and vanadium/calcium/alumina as catalysts. On all catalysts used for the dehydrocyclization, major reactions were fragmentation of MB and 1-(p-tolyl)-2-methylbutenes (MBes), dehydrogenation of MB, double bond transfer, hydrogenation and 1,6-cyclization of MBes. Minor reactions were 1,5-cyclization of MBes and methyl group fragmentation of 1,6- cyclization products. Esterification reactions of NPG were performed using three different carboxylic acids: propionic, isobutyric and 2-ethylhexanoic acid. Commercial heterogeneous gellular (Dowex 50WX2), macroreticular (Amberlyst 15) type resins and homogeneous para-toluene sulfonic acid were used as catalysts. At first NPG reacted with carboxylic acids to form corresponding monoester and water. Then monoester esterified with carboxylic acid to form corresponding diester. In disproportionation reaction two monoester molecules formed NPG and corresponding diester. All these three reactions can attain equilibrium. Concerning esterification, water was removed from the reactor in order to prevent backward reaction. Skeletal isomerization experiments of 1-pentene were performed over HZSM-22 catalyst. Isomerization reactions of three different kind were detected: double bond, cis-trans and skeletal isomerization. Minor side reaction were dimerization and fragmentation. Monomolecular and bimolecular reaction mechanisms for skeletal isomerization explained experimental results almost equally well. Pseudohomogeneous kinetic parameters of reactions 1 and 2 were estimated by usual least squares fitting. Concerning reactions 3 and 4 kinetic parameters were estimated by the leastsquares method, but also the possible cross-correlation and identifiability of parameters were determined using Markov chain Monte Carlo (MCMC) method. Finally using MCMC method, the estimation of model parameters and predictions were performed according to the Bayesian paradigm. According to the fitting results suggested reaction mechanisms explained experimental results rather well. When the possible cross-correlation and identifiability of parameters (Reactions 3 and 4) were determined using MCMC method, the parameters identified well, and no pathological cross-correlation could be seen between any parameter pair.

Kernel-Based Ranking. Methods for Learning and Performance Estimation

Machine learning provides tools for automated construction of predictive models in data intensive areas of engineering and science. The family of regularized kernel methods have in the recent years become one of the mainstream approaches to machine learning, due to a number of advantages the methods share. The approach provides theoretically well-founded solutions to the problems of under- and overfitting, allows learning from structured data, and has been empirically demonstrated to yield high predictive performance on a wide range of application domains. Historically, the problems of classification and regression have gained the majority of attention in the field. In this thesis we focus on another type of learning problem, that of learning to rank. In learning to rank, the aim is from a set of past observations to learn a ranking function that can order new objects according to how well they match some underlying criterion of goodness. As an important special case of the setting, we can recover the bipartite ranking problem, corresponding to maximizing the area under the ROC curve (AUC) in binary classification. Ranking applications appear in a large variety of settings, examples encountered in this thesis include document retrieval in web search, recommender systems, information extraction and automated parsing of natural language. We consider the pairwise approach to learning to rank, where ranking models are learned by minimizing the expected probability of ranking any two randomly drawn test examples incorrectly. The development of computationally efficient kernel methods, based on this approach, has in the past proven to be challenging. Moreover, it is not clear what techniques for estimating the predictive performance of learned models are the most reliable in the ranking setting, and how the techniques can be implemented efficiently. The contributions of this thesis are as follows. First, we develop RankRLS, a computationally efficient kernel method for learning to rank, that is based on minimizing a regularized pairwise least-squares loss. In addition to training methods, we introduce a variety of algorithms for tasks such as model selection, multi-output learning, and cross-validation, based on computational shortcuts from matrix algebra. Second, we improve the fastest known training method for the linear version of the RankSVM algorithm, which is one of the most well established methods for learning to rank. Third, we study the combination of the empirical kernel map and reduced set approximation, which allows the large-scale training of kernel machines using linear solvers, and propose computationally efficient solutions to cross-validation when using the approach. Next, we explore the problem of reliable cross-validation when using AUC as a performance criterion, through an extensive simulation study. We demonstrate that the proposed leave-pair-out cross-validation approach leads to more reliable performance estimation than commonly used alternative approaches. Finally, we present a case study on applying machine learning to information extraction from biomedical literature, which combines several of the approaches considered in the thesis. The thesis is divided into two parts. Part I provides the background for the research work and summarizes the most central results, Part II consists of the five original research articles that are the main contribution of this thesis.

ESTIMATION OF CORK PRODUCTION USINGAERIAL IMAGERY1

ABSTRACT Inventory and prediction of cork harvest over time and space is important to forest managers who must plan and organize harvest logistics (transport, storage, etc.). Common field inventory methods including the stem density, diameter and height structure are costly and generally point (plot) based. Furthermore, the irregular horizontal structure of cork oak stands makes it difficult, if not impossible, to interpolate between points. We propose a new method to estimate cork production using digital multispectral aerial imagery. We study the spectral response of individual trees in visible and near infrared spectra and then correlate that response with cork production prior to harvest. We use ground measurements of individual trees production to evaluate the model’s predictive capacity. We propose 14 candidate variables to predict cork production based on crown size in combination with different NDVI index derivates. We use Akaike Information Criteria to choose the best among them. The best model is composed of combinations of different NDVI derivates that include red, green, and blue channels. The proposed model is 15% more accurate than a model that includes only a crown projection without any spectral information.

Estimation of the potential evapotranspiration by a simplified penman method

The numerous methods for calculating the potential or reference evapotranspiration (ETo or ETP) almost always do for a 24-hour period, including values of climatic parameters throughout the nocturnal period (daily averages). These results have a nil effect on transpiration, constituting the main evaporative demand process in cases of localized irrigation. The aim of the current manuscript was to come up with a model rather simplified for the calculation of diurnal daily ETo. It deals with an alternative approach based on the theoretical background of the Penman method without having to consider values of aerodynamic conductance of latent and sensible heat fluxes, as well as data of wind speed and relative humidity of the air. The comparison between the diurnal values of ETo measured in weighing lysimeters with elevated precision and estimated by either the Penman-Monteith method or the Simplified-Penman approach in study also points out a fairly consistent agreement among the potential demand calculation criteria. The Simplified-Penman approach was a feasible alternative to estimate ETo under the local meteorological conditions of two field trials. With the availability of the input data required, such a method could be employed in other climatic regions for scheduling irrigation.

Comparison of maps of spatial variability of soil resistance to penetration constructed with and without covariables using a spatial linear model

A study about the spatial variability of data of soil resistance to penetration (RSP) was conducted at layers 0.0-0.1 m, 0.1-0.2 m and 0.2-0.3 m depth, using the statistical methods in univariate forms, i.e., using traditional geostatistics, forming thematic maps by ordinary kriging for each layer of the study. It was analyzed the RSP in layer 0.2-0.3 m depth through a spatial linear model (SLM), which considered the layers 0.0-0.1 m and 0.1-0.2 m in depth as covariable, obtaining an estimation model and a thematic map by universal kriging. The thematic maps of the RSP at layer 0.2-0.3 m depth, constructed by both methods, were compared using measures of accuracy obtained from the construction of the matrix of errors and confusion matrix. There are similarities between the thematic maps. All maps showed that the RSP is higher in the north region.

Construction and calibration of TDR probes for volumetric water content estimation in a Distroferric Red Latosol

The determination of volumetric water content of soils is an important factor in irrigation management. Among the indirect methods for estimating, the time-domain reflectometry (TDR) technique has received a significant attention. Like any other technique, it has advantages and disadvantages, but its greatest disadvantage is the need of calibration and high cost of acquisition. The main goal of this study was to establish a calibration model for the TDR equipment, Trase System Model 6050X1, to estimate the volumetric water content in a Distroferric Red Latosol. The calibration was carried out in a laboratory with disturbed soil samples under study, packed in PVC columns of a volume of 0.0078m³. The TDR probes were handcrafted with three rods and 0.20m long. They were vertically installed in soil columns, with a total of five probes per column and sixteen columns. The weightings were carried out in a digital scale, while daily readings of dielectric constant were obtained in TDR equipment. The linear model θν = 0.0103 Ka + 0.1900 to estimate the studied volumetric water content showed an excellent coefficient of determination (0.93), enabling the use of probes in indirect estimation of soil moisture.

Estimation methods of reference evapotranspiration (ETo) for Uberlândia -MG

One approach to verify the adequacy of estimation methods of reference evapotranspiration is the comparison with the Penman-Monteith method, recommended by the United Nations of Food and Agriculture Organization - FAO, as the standard method for estimating ET0. This study aimed to compare methods for estimating ET0, Makkink (MK), Hargreaves (HG) and Solar Radiation (RS), with Penman-Monteith (PM). For this purpose, we used daily data of global solar radiation, air temperature, relative humidity and wind speed for the year 2010, obtained through the automatic meteorological station, with latitude 18° 91' 66" S, longitude 48° 25' 05" W and altitude of 869m, at the National Institute of Meteorology situated in the Campus of Federal University of Uberlandia - MG, Brazil. Analysis of results for the period were carried out in daily basis, using regression analysis and considering the linear model y = ax, where the dependent variable was the method of Penman-Monteith and the independent, the estimation of ET0 by evaluated methods. Methodology was used to check the influence of standard deviation of daily ET0 in comparison of methods. The evaluation indicated that methods of Solar Radiation and Penman-Monteith cannot be compared, yet the method of Hargreaves indicates the most efficient adjustment to estimate ETo.