Multisensor change detection with nonlinear canonical correlation

The analysis of multi-modal and multi-sensor images is nowadays of paramount importance for Earth Observation (EO) applications. There exist a variety of methods that aim at fusing the different sources of information to obtain a compact representation of such datasets. However, for change detection existing methods are often unable to deal with heterogeneous image sources and very few consider possible nonlinearities in the data. Additionally, the availability of labeled information is very limited in change detection applications. For these reasons, we present the use of a semi-supervised kernel-based feature extraction technique. It incorporates a manifold regularization accounting for the geometric distribution and jointly addressing the small sample problem. An exhaustive example using Landsat 5 data illustrates the potential of the method for multi-sensor change detection.

Domain Adaptation in remote sensing: increasing the portability of land-cover classifiers

Among the types of remote sensing acquisitions, optical images are certainly one of the most widely relied upon data sources for Earth observation. They provide detailed measurements of the electromagnetic radiation reflected or emitted by each pixel in the scene. Through a process termed supervised land-cover classification, this allows to automatically yet accurately distinguish objects at the surface of our planet. In this respect, when producing a land-cover map of the surveyed area, the availability of training examples representative of each thematic class is crucial for the success of the classification procedure. However, in real applications, due to several constraints on the sample collection process, labeled pixels are usually scarce. When analyzing an image for which those key samples are unavailable, a viable solution consists in resorting to the ground truth data of other previously acquired images. This option is attractive but several factors such as atmospheric, ground and acquisition conditions can cause radiometric differences between the images, hindering therefore the transfer of knowledge from one image to another. The goal of this Thesis is to supply remote sensing image analysts with suitable processing techniques to ensure a robust portability of the classification models across different images. The ultimate purpose is to map the land-cover classes over large spatial and temporal extents with minimal ground information. To overcome, or simply quantify, the observed shifts in the statistical distribution of the spectra of the materials, we study four approaches issued from the field of machine learning. First, we propose a strategy to intelligently sample the image of interest to collect the labels only in correspondence of the most useful pixels. This iterative routine is based on a constant evaluation of the pertinence to the new image of the initial training data actually belonging to a different image. Second, an approach to reduce the radiometric differences among the images by projecting the respective pixels in a common new data space is presented. We analyze a kernel-based feature extraction framework suited for such problems, showing that, after this relative normalization, the cross-image generalization abilities of a classifier are highly increased. Third, we test a new data-driven measure of distance between probability distributions to assess the distortions caused by differences in the acquisition geometry affecting series of multi-angle images. Also, we gauge the portability of classification models through the sequences. In both exercises, the efficacy of classic physically- and statistically-based normalization methods is discussed. Finally, we explore a new family of approaches based on sparse representations of the samples to reciprocally convert the data space of two images. The projection function bridging the images allows a synthesis of new pixels with more similar characteristics ultimately facilitating the land-cover mapping across images.

Origin of the spectral shifts among the early intermediates of the rhodopsin photocycle.

A combined strategy based on the computation of absorption energies, using the ZINDO/S semiempirical method, for a statistically relevant number of thermally sampled configurations extracted from QM/MM trajectories is used to establish a one-to-one correspondence between the structures of the different early intermediates (dark, batho, BSI, lumi) involved in the initial steps of the rhodopsin photoactivation mechanism and their optical spectra. A systematic analysis of the results based on a correlation-based feature selection algorithm shows that the origin of the color shifts among these intermediates can be mainly ascribed to alterations in intrinsic properties of the chromophore structure, which are tuned by several residues located in the protein binding pocket. In addition to the expected electrostatic and dipolar effects caused by the charged residues (Glu113, Glu181) and to strong hydrogen bonding with Glu113, other interactions such as π-stacking with Ala117 and Thr118 backbone atoms, van der Waals contacts with Gly114 and Ala292, and CH/π weak interactions with Tyr268, Ala117, Thr118, and Ser186 side chains are found to make non-negligible contributions to the modulation of the color tuning among the different rhodopsin photointermediates.

Semi-supervised segmentation of ultrasound images based on patch representation and continuous min cut.

Ultrasound segmentation is a challenging problem due to the inherent speckle and some artifacts like shadows, attenuation and signal dropout. Existing methods need to include strong priors like shape priors or analytical intensity models to succeed in the segmentation. However, such priors tend to limit these methods to a specific target or imaging settings, and they are not always applicable to pathological cases. This work introduces a semi-supervised segmentation framework for ultrasound imaging that alleviates the limitation of fully automatic segmentation, that is, it is applicable to any kind of target and imaging settings. Our methodology uses a graph of image patches to represent the ultrasound image and user-assisted initialization with labels, which acts as soft priors. The segmentation problem is formulated as a continuous minimum cut problem and solved with an efficient optimization algorithm. We validate our segmentation framework on clinical ultrasound imaging (prostate, fetus, and tumors of the liver and eye). We obtain high similarity agreement with the ground truth provided by medical expert delineations in all applications (94% DICE values in average) and the proposed algorithm performs favorably with the literature.

An Efficient Segmentation Method for Ultrasound Images based on a Semi-supervised Approach and Patch-based Features

Segmenting ultrasound images is a challenging problemwhere standard unsupervised segmentation methods such asthe well-known Chan-Vese method fail. We propose in thispaper an efficient segmentation method for this class ofimages. Our proposed algorithm is based on asemi-supervised approach (user labels) and the use ofimage patches as data features. We also consider thePearson distance between patches, which has been shown tobe robust w.r.t speckle noise present in ultrasoundimages. Our results on phantom and clinical data show avery high similarity agreement with the ground truthprovided by a medical expert.

Association of serum homocysteine with major depressive disorder: results from a large population-based study.

BACKGROUND: Studies on the association between homocysteine levels and depression have shown conflicting results. To examine the association between serum total homocysteine (tHcy) levels and major depressive disorder (MDD) in a large community sample with an extended age range. METHODS: A total of 3392 men and women aged 35-66 years participating in the CoLaus study and its psychiatric arm (PsyCoLaus) were included in the analyses. High tHcy measured from fasting blood samples was defined as a concentration ≥15μmol/L. MDD was assessed using the semi-structured Diagnostic Interview for Genetics Studies. RESULTS: In multivariate analyses, elevated tHcy levels were associated with greater odds of meeting the diagnostic criteria for lifetime MDD among men (OR=1.71; 95% CI, 1.18-2.50). This was particularly the case for remitted MDD. Among women, there was no significant association between tHcy levels and MDD and the association tended to be in the opposite direction (OR=0.61; 95% CI, 0.34-1.08). CONCLUSIONS: In this large population-based study, elevated tHcy concentrations are associated with lifetime MDD and particularly with remitted MDD among men.

EEG-based functional brain networks: does the network size matter?

Functional connectivity in human brain can be represented as a network using electroencephalography (EEG) signals. These networks--whose nodes can vary from tens to hundreds--are characterized by neurobiologically meaningful graph theory metrics. This study investigates the degree to which various graph metrics depend upon the network size. To this end, EEGs from 32 normal subjects were recorded and functional networks of three different sizes were extracted. A state-space based method was used to calculate cross-correlation matrices between different brain regions. These correlation matrices were used to construct binary adjacency connectomes, which were assessed with regards to a number of graph metrics such as clustering coefficient, modularity, efficiency, economic efficiency, and assortativity. We showed that the estimates of these metrics significantly differ depending on the network size. Larger networks had higher efficiency, higher assortativity and lower modularity compared to those with smaller size and the same density. These findings indicate that the network size should be considered in any comparison of networks across studies.

Simulation-based systematic review of imatinib population pharmacokinetics and PK-PD relationships in chronic myeloid leukemia (CML) patients

Objectives: Several population pharmacokinetic (PPK) and pharmacokinetic-pharmacodynamic (PK-PD) analyses have been performed with the anticancer drug imatinib. Inspired by the approach of meta-analysis, we aimed to compare and combine results from published studies in a useful way - in particular for improving the clinical interpretation of imatinib concentration measurements in the scope of therapeutic drug monitoring (TDM). Methods: Original PPK analyses and PK-PD studies (PK surrogate: trough concentration Cmin; PD outcomes: optimal early response and specific adverse events) were searched systematically on MEDLINE. From each identified PPK model, a predicted concentration distribution under standard dosage was derived through 1000 simulations (NONMEM), after standardizing model parameters to common covariates. A "reference range" was calculated from pooled simulated concentrations in a semi-quantitative approach (without specific weighting) over the whole dosing interval. Meta-regression summarized relationships between Cmin and optimal/suboptimal early treatment response. Results: 9 PPK models and 6 relevant PK-PD reports in CML patients were identified. Model-based predicted median Cmin ranged from 555 to 1388 ng/ml (grand median: 870 ng/ml and inter-quartile range: 520-1390 ng/ml). The probability to achieve optimal early response was predicted to increase from 60 to 85% from 520 to 1390 ng/ml across PK-PD studies (odds ratio for doubling Cmin: 2.7). Reporting of specific adverse events was too heterogeneous to perform a regression analysis. The general frequency of anemia, rash and fluid retention increased however consistently with Cmin, but less than response probability. Conclusions: Predicted drug exposure may differ substantially between various PPK analyses. In this review, heterogeneity was mainly attributed to 2 "outlying" models. The established reference range seems to cover the range where both good efficacy and acceptable tolerance are expected for most patients. TDM guided dose adjustment appears therefore justified for imatinib in CML patients. Its usefulness remains now to be prospectively validated in a randomized trial.

Predictors of nonresponse in a questionnaire-based outcome study of vocational rehabilitation patients.

OBJECTIVE: To identify predictors of nonresponse to a self-report study of patients with orthopedic trauma hospitalized for vocational rehabilitation between November 15, 2003, and December 31, 2005. The role of biopsychosocial complexity, assessed using the INTERMED, was of particular interest. DESIGN: Cohort study. Questionnaires with quality of life, sociodemographic, and job-related questions were given to patients at hospitalization and 1 year after discharge. Sociodemographic data, biopsychosocial complexity, and presence of comorbidity were available at hospitalization (baseline) for all eligible patients. Logistic regression models were used to test a number of baseline variables as potential predictors of nonresponse to the questionnaires at each of the 2 time points. SETTING: Rehabilitation clinic. PARTICIPANTS: Patients (N=990) hospitalized for vocational rehabilitation over a period of 2 years. INTERVENTIONS: Not applicable. MAIN OUTCOME MEASURE: Nonresponse to the questionnaires was the binary dependent variable. RESULTS: Patients with high biopsychosocial complexity, foreign native language, or low educational level were less likely to respond at both time points. Younger patients were less likely to respond at 1 year. Those living in a stable partnership were less likely than singles to respond at hospitalization. Sex, psychiatric, and somatic comorbidity and alcoholism were never associated with nonresponse. CONCLUSIONS: We stress the importance of assessing biopsychosocial complexity to predict nonresponse. Furthermore, the factors we found to be predictive of nonresponse are also known to influence treatment outcome and vocational rehabilitation. Therefore, it is important to increase the response rate of the groups of concern in order to reduce selection bias in epidemiologic investigations.

Characterization of tumor antigen specific CD8+ T cells and semi-invariant Vα24/Vβ11 NKT cells in tumor invaded liver

Summary: Detailed knowledge on tumor antigen expression and specific immune cells is required for a rational design of immunotherapy for patients with tumor invaded liver. In this study, we confirmed that Cancer/Testis (CT) tumor-associated antigens are frequently expressed in hepatocellular carcinoma (HCC) and searched for the presence of CD8+ T cells specific for these antigens. In 2/10 HLA-A2+ patients with HCC, we found that MAGE-A10 and/or SSX-2 specific CD8+ T cells naturally responded to the disease, since they were enriched in tumor lesions but not in non-tumoral liver. Isolated T cells specifically and strongly killed tumor cells in vitro, suggesting that these CTL were selected in vivo for high avidity antigen recognition, providing the rational for specific immunotherapy of HCC, based on immunization with CT antigens such as MAGE-Al 0 and SSX-2. Type 1 NKT cells express an invariant TCR α chain (Vα24.1α18, paired with Vβ11 in human) and share a specific reactivity to αGalactosylceramide (αGC) presented by CD1d. These cells can display paradoxical immuno-regulatory properties including strong anti-tumor effects upon αGC administration in murine models. To understand why NKT cells were not sufficiently protective against tumor development in patients with tumor invaded liver, we characterized the diversity of Vα24/Vβ11 NKT cells in healthy donors (HD) and cancer patients: NKT cells from HD and patients were generally diverse in terms of TCR β chain (Vβ11) variability and NKT cells from HD showed a variable recognition of αGC loaded CD 1 d multimers. Vα24/ Vβ11 NKT cells can be divided in 3 populations, the CD4, DN (CD4-/CD8-) and CD8 NKT cell subsets that show distinct ability of cytokine production. In addition, our functional analysis revealed that DN and CD8 subsets displayed a higher cytolytic potential and a weaker IFNγ release than the CD4 NKT cell subset. NKT cell subsets were variably represented in the blood of HD and cancer patients. However, HD with high NKT cell frequencies displayed an enrichment of the DN and CD8 subsets, and few of them were suggestive of an oligoclonal expansion in vivo. Comparable NKT cell frequencies were found between blood, non-tumoral liver and tumor of patients. In contrast, we identified a gradual enrichment of CD4 NKT cells from blood to the liver and to the tumor, together with a decrease of DN and CD8 NKT cell subsets. Most patient derived NKT cells were unresponsive upon αGalactosylceramide stimulation ex vivo; NKT cells from few patients displayed a weak responsiveness with different cytokine polarization. The NKT cell repertoire was thus different in tumor tissue, suggesting that CD4 NKT cells infiltrating tumors may be detrimental for protection against tumors and instead may favour the tumor growth/recurrence as recently reported in mice. Résumé en français scientifique : Afin de développer le traitement des patients porteurs d'une tumeur dans le foie par immunothérapie, de nouvelles connaissances sont requises concernant l'expression d'antigènes par les tumeurs et les cellules immunitaires spécifiques de ces antigènes. Nous avons vérifié que des antigènes associés aux tumeurs, tels que les antigènes « Cancer-Testis » (CT), sont fréquemment exprimés par le carcinome hepatocéllulaire (CHC). La recherche de lymphocytes T CD8+ spécifiques (CTL) de ces antigènes a révélé que des CTL spécifiques de MAGE-A10 et/ou SSX-2 ont répondu naturellement à la tumeur chez 2/10 patients étudiés. Ces cellules étaient présentes dans les lésions tumorales mais pas dans le foie adjacent. De plus, ces CTL ont démontré une activité cytolytique forte et spécifique contre les cellules tumorales in vitro, ce qui suggère que ces CTL ont été sélectionnés pour une haute avidité de reconnaissance de l'antigène in vivo. Ces données fournissent une base pour l'immunothérapie spécifique du CHC, en proposant de cibler les antigènes CT tels que MAGE-A10 ou SSX-2. Les cellules NKT de type 1 ont une chaîne α de TCR qui est invariante (chez l'homme, Vα24Jα18, apparié avec Vβ11) et reconnaissent spécifiquement l'αGalactosylceramide (αGC) présenté par CD1d. Ces cellules ont des propriétés immuno¬régulatrices qui peuvent être parfois contradictoires et leur activation par l'αGC induit une forte protection anti-tumorale chez la souris: Afin de comprendre pourquoi ces cellules ne sont pas assez protectrices contre le développement des tumeurs dans le foie chez l'homme, nous avons étudié la diversité des cellules NKT Vα24/Vβ11 d'individus sains (IS) et de patients cancéreux. Les cellules NKT peuvent être sous-divisées en 3 populations : Les CD4, DN (CD4- /CD8-) ou CDS, qui ont la capacité de produire des cytokines différentes. Nos analyses fonctionnelles ont aussi révélé que les sous-populations DN et CD8 ont un potentiel cytolytique plus élevé et une production d'IFNγ plus faible que la sous-population CD4. Ces sous-populations sont représentées de manière variable dans le sang des IS ou des patients. Cependant, les IS avec un taux élevé de cellules NKT ont un enrichissement des sous- populations DN ou CDS, et certains suggèrent qu'il s'agit d'une expansion oligo-clonale in vivo. Les patients avaient des fréquences comparables de cellules NKT entre le sang, le foie et la tumeur. Par contre, la sous-population CD4 était progressivement enrichie du sang vers le foie et la tumeur, tandis que les sous-populations DN ou CD8 était perdues. La plupart des cellules NKT des patients ne réagissaient pas lors de stimulation avec l'αGC ex vivo et les cellules NKT de quelques patients répondaient faiblement et avec des polarisations de cytokines différentes. Ces données suggèrent que les cellules NKT CD4, prédominantes dans les tumeurs, sont inefficaces pour la lutte anti-tumorale et pourraient même favoriser la croissance ou la récurrence tumorale. Donc, une mobilisation spécifique des cellules NKT CD4 négatives par immunothérapie pourrait favoriser l'immunité contre des tumeurs chez l'homme. Résumé en français pour un large public Au sein des globules blancs, les lymphocytes T expriment un récepteur (le TCR), qui est propre à chacun d'entre eux et leur permet d'accrocher de manière très spécifique une molécule appelée antigène. Ce TCR est employé par les lymphocytes pour inspecter les antigènes associés avec des molécules présentatrices à la surface des autres cellules. Les lymphocytes T CD8 reconnaissent un fragment de protéine (ou peptide), qui est présenté par une des molécules du Complexe Majeur d'Histocompatibilité de classe I et tuent la cellule qui présente ce peptide. Ils sont ainsi bien adaptés pour éliminer les cellules qui présentent un peptide issu d'un virus quand la cellule est infectée. D'autres cellules T CD8 reconnaissent des peptides comme les antigènes CT, qui sont produits anormalement par les cellules cancéreuses. Nous avons confirmé que les antigènes CT sont fréquemment exprimés par le cancer du foie. Nous avons également identifié des cellules T CD8 spécifiques d'antigènes CT dans la tumeur, mais pas dans le foie normal de 2 patients sur 10. Cela signifie que ces lymphocytes peuvent être naturellement activés contre la tumeur et sont capables de la trouver. De plus les lymphocytes issus d'un patient ont démontré une forte sensibilité pour reconnaître l'antigène et tuent spécifiquement les cellules tumorales. Les antigènes CT représentent donc des cibles intéressantes qui pourront être intégrés dans des vaccins thérapeutiques du cancer du foie. De cette manière, les cellules T CD8 du patient lui-même pourront être induites à détruire de manière spécifique les cellules cancéreuses. Un nouveau type de lymphocytes T a été récemment découvert: les lymphocytes NKT. Quand ils reconnaissent un glycolipide présenté par la molécule CD1d, ils sont capables, de manière encore incomprise, d'initier, d'augmenter, ou à l'inverse d'inhiber la défense immunitaire. Ces cellules NKT ont démontré qu'elles jouent un rôle important dans la défense contre les tumeurs et particulièrement dans le foie des souris. Nous avons étudié les cellules NKT de patients atteints d'une tumeur dans le foie, afin de comprendre pourquoi elles ne sont pas assez protectrice chez l'homme. Les lymphocytes NKT peuvent être sous-divisés en 3 populations: Les CD4, les DN (CD4-/CD8-) et les CD8. Ces 3 classes de NKT peuvent produire différents signaux chimiques appelés cytokines. Contrairement aux cellules NKT DN ou CDS, seules les cellules NKT CD4 sont capables de produire des cytokines qui sont défavorables pour la défense anti-tumorale. Par ailleurs nous avons trouvé que les cellules NKT CD4 tuent moins bien les cellules cancéreuses que les cellules NKT DN ou CD8. L'analyse des cellules NKT, fraîchement extraites du sang, du foie et de la tumeur de patients a révélé que les cellules NKT CD4 sont progressivement enrichies du sang vers le foie et la tumeur. La large prédominance des NKT CD4 à l'intérieur des tumeurs suggère que, chez l'homme, ces cellules sont inappropriées pour la lutte anti-tumorale. Par ailleurs, la plupart des cellules NKT de patients n'étaient pas capables de produire des cytokines après stimulation avec un antigène. Cela explique également pourquoi ces cellules ne protègent pas contre les tumeurs dans le foie.

Binary classification of dyslipidemia from the waist-to-hip ratio and body mass index: a comparison of linear, logistic, and CART models.

BACKGROUND: We sought to improve upon previously published statistical modeling strategies for binary classification of dyslipidemia for general population screening purposes based on the waist-to-hip circumference ratio and body mass index anthropometric measurements. METHODS: Study subjects were participants in WHO-MONICA population-based surveys conducted in two Swiss regions. Outcome variables were based on the total serum cholesterol to high density lipoprotein cholesterol ratio. The other potential predictor variables were gender, age, current cigarette smoking, and hypertension. The models investigated were: (i) linear regression; (ii) logistic classification; (iii) regression trees; (iv) classification trees (iii and iv are collectively known as "CART"). Binary classification performance of the region-specific models was externally validated by classifying the subjects from the other region. RESULTS: Waist-to-hip circumference ratio and body mass index remained modest predictors of dyslipidemia. Correct classification rates for all models were 60-80%, with marked gender differences. Gender-specific models provided only small gains in classification. The external validations provided assurance about the stability of the models. CONCLUSIONS: There were no striking differences between either the algebraic (i, ii) vs. non-algebraic (iii, iv), or the regression (i, iii) vs. classification (ii, iv) modeling approaches. Anticipated advantages of the CART vs. simple additive linear and logistic models were less than expected in this particular application with a relatively small set of predictor variables. CART models may be more useful when considering main effects and interactions between larger sets of predictor variables.

Parts, places, and perspectives : a theory of spatial relations based an mereotopology and convexity

This thesis suggests to carry on the philosophical work begun in Casati's and Varzi's seminal book Parts and Places, by extending their general reflections on the basic formal structure of spatial representation beyond mereotopology and absolute location to the question of perspectives and perspective-dependent spatial relations. We show how, on the basis of a conceptual analysis of such notions as perspective and direction, a mereotopological theory with convexity can express perspectival spatial relations in a strictly qualitative framework. We start by introducing a particular mereotopological theory, AKGEMT, and argue that it constitutes an adequate core for a theory of spatial relations. Two features of AKGEMT are of particular importance: AKGEMT is an extensional mereotopology, implying that sameness of proper parts is a sufficient and necessary condition for identity, and it allows for (lower- dimensional) boundary elements in its domain of quantification. We then discuss an extension of AKGEMT, AKGEMTS, which results from the addition of a binary segment operator whose interpretation is that of a straight line segment between mereotopological points. Based on existing axiom systems in standard point-set topology, we propose an axiomatic characterisation of the segment operator and show that it is strong enough to sustain complex properties of a convexity predicate and a convex hull operator. We compare our segment-based characterisation of the convex hull to Cohn et al.'s axioms for the convex hull operator, arguing that our notion of convexity is significantly stronger. The discussion of AKGEMTS defines the background theory of spatial representation on which the developments in the second part of this thesis are built. The second part deals with perspectival spatial relations in two-dimensional space, i.e., such relations as those expressed by 'in front of, 'behind', 'to the left/right of, etc., and develops a qualitative formalism for perspectival relations within the framework of AKGEMTS. Two main claims are defended in part 2: That perspectival relations in two-dimensional space are four- place relations of the kind R(x, y, z, w), to be read as x is i?-related to y as z looks at w; and that these four-place structures can be satisfactorily expressed within the qualitative theory AKGEMTS. To defend these two claims, we start by arguing for a unified account of perspectival relations, thus rejecting the traditional distinction between 'relative' and 'intrinsic' perspectival relations. We present a formal theory of perspectival relations in the framework of AKGEMTS, deploying the idea that perspectival relations in two-dimensional space are four-place relations, having a locational and a perspectival part and show how this four-place structure leads to a unified framework of perspectival relations. Finally, we present a philosophical motivation to the idea that perspectival relations are four-place, cashing out the thesis that perspectives are vectorial properties and argue that vectorial properties are relations between spatial entities. Using Fine's notion of "qua objects" for an analysis of points of view, we show at last how our four-place approach to perspectival relations compares to more traditional understandings.

Imputation in data fusion of heterogeneous data sets a model-based numerical experiment

Given the very large amount of data obtained everyday through population surveys, much of the new research again could use this information instead of collecting new samples. Unfortunately, relevant data are often disseminated into different files obtained through different sampling designs. Data fusion is a set of methods used to combine information from different sources into a single dataset. In this article, we are interested in a specific problem: the fusion of two data files, one of which being quite small. We propose a model-based procedure combining a logistic regression with an Expectation-Maximization algorithm. Results show that despite the lack of data, this procedure can perform better than standard matching procedures.

Uncertainty Quantification Of A Semi-Supervised Support Vector Regression Reservoir Model

Uncertainty quantification of petroleum reservoir models is one of the present challenges, which is usually approached with a wide range of geostatistical tools linked with statistical optimisation or/and inference algorithms. Recent advances in machine learning offer a novel approach to model spatial distribution of petrophysical properties in complex reservoirs alternative to geostatistics. The approach is based of semisupervised learning, which handles both ?labelled? observed data and ?unlabelled? data, which have no measured value but describe prior knowledge and other relevant data in forms of manifolds in the input space where the modelled property is continuous. Proposed semi-supervised Support Vector Regression (SVR) model has demonstrated its capability to represent realistic geological features and describe stochastic variability and non-uniqueness of spatial properties. On the other hand, it is able to capture and preserve key spatial dependencies such as connectivity of high permeability geo-bodies, which is often difficult in contemporary petroleum reservoir studies. Semi-supervised SVR as a data driven algorithm is designed to integrate various kind of conditioning information and learn dependences from it. The semi-supervised SVR model is able to balance signal/noise levels and control the prior belief in available data. In this work, stochastic semi-supervised SVR geomodel is integrated into Bayesian framework to quantify uncertainty of reservoir production with multiple models fitted to past dynamic observations (production history). Multiple history matched models are obtained using stochastic sampling and/or MCMC-based inference algorithms, which evaluate posterior probability distribution. Uncertainty of the model is described by posterior probability of the model parameters that represent key geological properties: spatial correlation size, continuity strength, smoothness/variability of spatial property distribution. The developed approach is illustrated with a fluvial reservoir case. The resulting probabilistic production forecasts are described by uncertainty envelopes. The paper compares the performance of the models with different combinations of unknown parameters and discusses sensitivity issues.