A Survey of Active Learning Algorithms for Supervised Remote Sensing Image Classification

Defining an efficient training set is one of the most delicate phases for the success of remote sensing image classification routines. The complexity of the problem, the limited temporal and financial resources, as well as the high intraclass variance can make an algorithm fail if it is trained with a suboptimal dataset. Active learning aims at building efficient training sets by iteratively improving the model performance through sampling. A user-defined heuristic ranks the unlabeled pixels according to a function of the uncertainty of their class membership and then the user is asked to provide labels for the most uncertain pixels. This paper reviews and tests the main families of active learning algorithms: committee, large margin, and posterior probability-based. For each of them, the most recent advances in the remote sensing community are discussed and some heuristics are detailed and tested. Several challenging remote sensing scenarios are considered, including very high spatial resolution and hyperspectral image classification. Finally, guidelines for choosing the good architecture are provided for new and/or unexperienced user.

Direct analysis of peptide binding to cell-associated MHC class I molecules.

Exogenously added synthetic peptides can mimic endogenously produced antigenic peptides recognized on target cells by MHC class I-restricted cytolytic T lymphocytes. While it is assumed that exogenous peptides associate with class I molecules on the target cell surface, direct binding of peptides to cell-associated class I molecules has been difficult to demonstrate. Using a newly developed binding assay based on photoaffinity labeling, we have investigated the interaction of two antigenic peptides, known to be recognized in the context of H-2Kd or H-2Db, respectively, with 20 distinct class I alleles on living cells. None of the class I alleles tested, with the exception of H-2Kd or H-2Db, bound either of the peptides, thus demonstrating the exquisite specificity of peptide binding to class I molecules. Moreover, peptide binding to cell-associated H-2Kd was drastically reduced when metabolic energy, de novo protein synthesis or protein egress from the endoplasmic reticulum was inhibited. It is thus likely that exogenously added peptides do not associate with the bulk of class I molecules expressed at the cell surface, but rather bind to short-lived molecules devoid of endogenous peptides.

Structured output SVM for remote sensing image classification

In the recent years, kernel methods have revealed very powerful tools in many application domains in general and in remote sensing image classification in particular. The special characteristics of remote sensing images (high dimension, few labeled samples and different noise sources) are efficiently dealt with kernel machines. In this paper, we propose the use of structured output learning to improve remote sensing image classification based on kernels. Structured output learning is concerned with the design of machine learning algorithms that not only implement input-output mapping, but also take into account the relations between output labels, thus generalizing unstructured kernel methods. We analyze the framework and introduce it to the remote sensing community. Output similarity is here encoded into SVM classifiers by modifying the model loss function and the kernel function either independently or jointly. Experiments on a very high resolution (VHR) image classification problem shows promising results and opens a wide field of research with structured output kernel methods.

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.

Noninvasive imaging of 5-HT3 receptor trafficking in live cells: from biosynthesis to endocytosis.

Sequential stages in the life cycle of the ionotropic 5-HT(3) receptor (5-HT(3)R) were resolved temporally and spatially in live cells by multicolor fluorescence confocal microscopy. The insertion of the enhanced cyan fluorescent protein into the large intracellular loop delivered a fluorescent 5-HT(3)R fully functional in terms of ligand binding specificity and channel activity, which allowed for the first time a complete real-time visualization and documentation of intracellular biogenesis, membrane targeting, and ligand-mediated internalization of a receptor belonging to the ligand-gated ion channel superfamily. Fluorescence signals of newly expressed receptors were detectable in the endoplasmic reticulum about 3 h after transfection onset. At this stage receptor subunits assembled to form active ligand binding sites as demonstrated in situ by binding of a fluorescent 5-HT(3)R-specific antagonist. After novel protein synthesis was chemically blocked, the 5-HT(3) R populations in the endoplasmic reticulum and Golgi cisternae moved virtually quantitatively to the cell surface, indicating efficient receptor folding and assembly. Intracellular 5-HT(3) receptors were trafficking in vesicle-like structures along microtubules to the cell surface at a velocity generally below 1 mum/s and were inserted into the plasma membrane in a characteristic cluster distribution overlapping with actin-rich domains. Internalization of cell surface 5-HT(3) receptors was observed within minutes after exposure to an extracellular agonist. Our orchestrated use of spectrally distinguishable fluorescent labels for the receptor, its cognate ligand, and specific organelle markers can be regarded as a general approach allowing subcellular insights into dynamic processes of membrane receptor trafficking.

Photoaffinity labeling of the T cell receptor on living cytotoxic T lymphocytes.

Using a direct binding assay based on photoaffinity labeling, we have studied the interaction of an antigenic peptide with MHC class I molecules and the TCR on living cells. Two photoreactive derivatives of the H-2Kd (Kd) restricted Plasmodium berghei circumsporozoite (PbCS) peptide 253-260 (YIPSAEKI) were used. The first derivative contained an N-terminal photoreactive iodo, 4-azido salicyloyl (IASA) group and biotin on the TCR contact residue Lys259 [IASA-YIPSAEK(biotin)I]. As previously described, this derivative selectively bound to and labeled the Kd molecule. The second photoreactive compound, the isomeric biotin-YIPSAEK(IASA)I, also efficiently bound to the Kd molecule, but failed to label this protein. A CTL clone derived from a mouse immunized with this derivative recognized this conjugate but not the parental P. berghei circumsporozoite peptide or the [IASA-YIPSAEK-(biotin)I] derivative in an Kd-restricted manner. Incubation of the cloned CTL cells with biotin-YIPSAEK(IASA)I, but not its isomer, followed by UV irradiation resulted in photoaffinity labeling of the TCR-alpha chain that was dependent on the conjugate binding to the Kd molecule. The TCR labeling was partially inhibited by anti-LFA 1 and anti-ICAM1 mAb, but was increased by addition of beta 2m or soluble KdQ10. The exquisite labeling selectivity of the two photoprobes opens a new, direct approach to the molecular analysis of antigen presentation and recognition by living CTL.

Spin-labeling coronary MR angiography with steady-state free precession and radial k-space sampling: initial results in healthy volunteers.

The purpose of this study was to prospectively compare free-breathing navigator-gated cardiac-triggered three-dimensional steady-state free precession (SSFP) spin-labeling coronary magnetic resonance (MR) angiography performed by using Cartesian k-space sampling with that performed by using radial k-space sampling. A new dedicated placement of the two-dimensional selective labeling pulse and an individually adjusted labeling delay time approved by the institutional review board were used. In 14 volunteers (eight men, six women; mean age, 28.8 years) who gave informed consent, signal-to-noise ratio (SNR), contrast-to-noise ratio (CNR), vessel sharpness, vessel length, and subjective image quality were investigated. Differences between groups were analyzed with nonparametric tests (Wilcoxon, Pearson chi2). Radial imaging, as compared with Cartesian imaging, resulted in a significant reduction in the severity of motion artifacts, as well as an increase in SNR (26.9 vs 12.0, P < .05) in the coronary arteries and CNR (23.1 vs 8.8, P < .05) between the coronary arteries and the myocardium. A tendency toward improved vessel sharpness and vessel length was also found with radial imaging. Radial SSFP imaging is a promising technique for spin-labeling coronary MR angiography.

Phenotypic features of carbohydrate sulfotransferase 3 (CHST3) deficiency in 24 patients: congenital dislocations and vertebral changes as principal diagnostic features.

We recently reported on the deficiency of carbohydrate sulfotransferase 3 (CHST3; chondroitin-6-sulfotransferase) in six subjects diagnosed with recessive Larsen syndrome or humero-spinal dysostosis [Hermanns et al. (2008); Am J Hum Genet 82:1368-1374]. Since then, we have identified 17 additional families with CHST3 mutations and we report here on a series of 24 patients in 23 families. The diagnostic hypothesis prior to molecular analysis had been: Larsen syndrome (15 families), humero-spinal dysostosis (four cases), chondrodysplasia with multiple dislocations (CDMD "Megarbane type"; two cases), Desbuquois syndrome (one case), and spondylo-epiphyseal dysplasia (one case). In spite of the different diagnostic labels, the clinical features in these patients were similar and included dislocation of the knees and/or hips at birth, clubfoot, elbow joint dysplasia with subluxation and limited extension, short stature, and progressive kyphosis developing in late childhood. The most useful radiographic clues were the changes of the lumbar vertebrae. Twenty-four different CHST3 mutations were identified; 16 patients had homozygous mutations. We conclude that CHST3 deficiency presents at birth with congenital dislocations of knees, hips, and elbows, and is often diagnosed initially as Larsen syndrome, humero-spinal dysostosis, or chondrodysplasia with dislocations. The incidence of CHST3 deficiency seems to be higher than assumed so far. The clinical and radiographic pattern (joint dislocations, vertebral changes, normal carpal age, lack of facial flattening, and recessive inheritance) is characteristic and distinguishes CHST3 deficiency from other disorders with congenital dislocations such as filamin B-associated dominant Larsen syndrome and Desbuquois syndrome.

Alcohol drinking and cardiovascular risk in a population with high mean alcohol consumption.

Moderate alcohol consumption has been associated with lower coronary artery disease (CAD) risk. However, data on the CAD risk associated with high alcohol consumption are conflicting. The aim of this study was to examine the impact of heavier drinking on 10-year CAD risk in a population with high mean alcohol consumption. In a population-based study of 5,769 adults (aged 35 to 75 years) without cardiovascular disease in Switzerland, 1-week alcohol consumption was categorized as 0, 1 to 6, 7 to 13, 14 to 20, 21 to 27, 28 to 34, and > or =35 drinks/week or as nondrinkers (0 drinks/week), moderate (1 to 13 drinks/week), high (14 to 34 drinks/week), and very high (> or =35 drinks/week). Blood pressure and lipids were measured, and 10-year CAD risk was calculated according to the Framingham risk score. Seventy-three percent (n = 4,214) of the participants consumed alcohol; 16% (n = 909) were high drinkers and 2% (n = 119) very high drinkers. In multivariate analysis, increasing alcohol consumption was associated with higher high-density lipoprotein cholesterol (from a mean +/- SE of 1.57 +/- 0.01 mmol/L in nondrinkers to 1.88 +/- 0.03 mmol/L in very high drinkers); triglycerides (1.17 +/- 1.01 to 1.32 +/- 1.05 mmol/L), and systolic and diastolic blood pressure (127.4 +/- 0.4 to 132.2 +/- 1.4 mm Hg and 78.7 +/- 0.3 to 81.7 +/- 0.9 mm Hg, respectively) (all p values for trend <0.001). Ten-year CAD risk increased from 4.31 +/- 0.10% to 4.90 +/- 0.37% (p = 0.03) with alcohol use, with a J-shaped relation. Increasing wine consumption was more related to high-density lipoprotein cholesterol levels, whereas beer and spirits were related to increased triglyceride levels. In conclusion, as measured by 10-year CAD risk, the protective effect of alcohol consumption disappears in very high drinkers, because the beneficial increase in high-density lipoprotein cholesterol is offset by the increases in blood pressure levels.

Differential roles of T cell receptor alpha and beta chains in ligand binding among H-2Kd-restricted cytolytic T lymphocyte clones specific for a photoreactive Plasmodium berghei circumsporozoite peptide derivative.

To study the interaction of T cell receptor with its ligand, a complex of a major histocompatibility complex molecule and a peptide, we derived H-2Kd-restricted cytolytic T lymphocyte clones from mice immunized with a Plasmodium berghei circumsporozoite peptide (PbCS) 252-260 (SYIPSAEKI) derivative containing photoreactive Nepsilon-[4-azidobenzoyl] lysine in place of Pro-255. This residue and Lys-259 were essential parts of the epitope recognized by these clones. Most of the clones expressed BV1S1A1 encoded beta chains along with specific complementary determining region (CDR) 3beta regions but diverse alpha chain sequences. Surprisingly, all T cell receptors were preferentially photoaffinity labeled on the alpha chain. For a representative T cell receptor, the photoaffinity labeled site was located in the Valpha C-strand. Computer modeling suggested the presence of a hydrophobic pocket, which is formed by parts of the Valpha/Jalpha C-, F-, and G-strands and adjacent CDR3alpha residues and structured to be able to avidly bind the photoreactive ligand side chain. We previously found that a T cell receptor specific for a PbCS peptide derivative containing this photoreactive side chain in position 259 similarly used a hydrophobic pocket located between the junctional CDR3 loops. We propose that this nonpolar domain in these locations allow T cell receptors to avidly and specifically bind epitopes containing non-peptidic side chains.

Hydrophobic labeling of (Na+,K+)-ATPase: further evidence that the beta subunit is embedded in the membrane bilayer.

O-Hexanoyl-3,5-diiodo-N-(4-azido-2-nitro-phenyl)tyramine has been used after photochemical conversion into the reactive nitrene to label (Na+,K+)-ATPase from Bufo marinus toad kidney. Immunochemical evidence indicates that the reagent labels both subunits of the enzyme in partially purified form as well as in microsomal membranes. These results support the view that the glycoprotein subunit, like the catalytic subunit, possesses hydrophobic domains by which it is integrated into the plasma membrane.

Effect of biological surlie ageing on the sparkling wines’ quality and elaoration

Report for the scientific sojourn at the Université de Bourgogne, France, from July until October 2007..Surlie ageing after second fermentation is a fundamental operation in the production of quality sparkling wine like Cava and Champagne. Recently, the importance of the interaction between wine and lees cell surface has been reported. Cell surface properties depending on wall biochemical composition are major determinants in microbial interactions, having important repercussions in several technological aspects. Sorption and flocculation are especially important in sparkling wine production, and are governed by distinct cell surface properties. The aim of the present research carried out during the four months of the stage was to know the implication of lees surface modifications occurring during surlie ageing in sparkling wine quality and elaboration. The relationship between physico-chemical properties such as hydrophobicity, charge and electron-donor characteristics, and the yeast surface sorption capacities, we determined these factors in a model system. Then, real industrial lees samples were investigated. The surface properties of sparkling wine lees from the same strain of Saccharomyces cerevisiae were characterized according to the time of surlie ageing, and their possible influence on lees sorption and flocculation capacity was evaluated. Surlie ageing after second fermentation is a fundamental operation in the production of quality sparkling wine like Cava and Champagne. Recently, the importance of the interaction between wine and lees cell surface has been reported. Cell surface properties depending on wall biochemical composition are major determinants in microbial interactions, having important repercussions in several technological aspects. Sorption and flocculation are especially important in sparkling wine production, and are governed by distinct cell surface properties. The aim of the present research carried out during the four months of the stage was to know the implication of lees surface modifications occurring during surlie ageing in sparkling wine quality and elaboration. The relationship between physico-chemical properties such as hydrophobicity, charge and electron-donor characteristics, and the yeast surface sorption capacities, we determined these factors in a model system. Then, real industrial lees samples were investigated. The surface properties of sparkling wine lees from the same strain of Saccharomyces cerevisiae were characterized according to the time of surlie ageing, and their possible influence on lees sorption and flocculation capacity was evaluated.

Sound Texture Modelling

In the PhD thesis “Sound Texture Modeling” we deal with statistical modelling or textural sounds like water, wind, rain, etc. For synthesis and classification. Our initial model is based on a wavelet tree signal decomposition and the modeling of the resulting sequence by means of a parametric probabilistic model, that can be situated within the family of models trainable via expectation maximization (hidden Markov tree model ). Our model is able to capture key characteristics of the source textures (water, rain, fire, applause, crowd chatter ), and faithfully reproduces some of the sound classes. In terms of a more general taxonomy of natural events proposed by Graver, we worked on models for natural event classification and segmentation. While the event labels comprise physical interactions between materials that do not have textural propierties in their enterity, those segmentation models can help in identifying textural portions of an audio recording useful for analysis and resynthesis. Following our work on concatenative synthesis of musical instruments, we have developed a pattern-based synthesis system, that allows to sonically explore a database of units by means of their representation in a perceptual feature space. Concatenative syntyhesis with “molecules” built from sparse atomic representations also allows capture low-level correlations in perceptual audio features, while facilitating the manipulation of textural sounds based on their physical and perceptual properties. We have approached the problem of sound texture modelling for synthesis from different directions, namely a low-level signal-theoretic point of view through a wavelet transform, and a more high-level point of view driven by perceptual audio features in the concatenative synthesis setting. The developed framework provides unified approach to the high-quality resynthesis of natural texture sounds. Our research is embedded within the Metaverse 1 European project (2008-2011), where our models are contributting as low level building blocks within a semi-automated soundscape generation system.

HLA photoaffinity labeling reveals overlapping binding of homologous melanoma-associated gene peptides by HLA-A1, HLA-A29, and HLA-B44.

Melanoma-associated genes (MAGEs) encode tumor-specific antigens that can be recognized by CD8+ cytotoxic T lymphocytes. To investigate the interaction of the HLA-A1-restricted MAGE-1 peptide 161-169 (EADPT-GHSY) with HLA class I molecules, photoreactive derivatives were prepared by single amino acid substitution with N beta-[iodo-4-azidosalicyloyl]-L-2,3-diaminopropionic acid. These derivatives were tested for their ability to bind to, and to photoaffinity-label, HLA-A1 on C1R.A1 cells. Only the derivatives containing the photoreactive amino acid in position 1 or 7 fulfilled both criteria. Testing the former derivative on 14 lymphoid cell lines expressing over 44 different HLA class I molecules indicated that it efficiently photoaffinity-labeled not only HLA-A1, but possibility also HLA-A29 and HLA-B44. MAGE peptide binding by HLA-A29 and HLA-B44 was confirmed by photoaffinity labeling with photoreactive MAGE-3 peptide derivatives on C1R.A29 and C1R.B44 cells, respectively. The different photoaffinity labeling systems were used to access the ability of the homologous peptides derived from MAGE-1, -2, -3, -4a, -4b, -6, and -12 to bind to HLA-A1, HLA-A29, and HLA-B44. All but the MAGE-2 and MAGE-12 nonapeptides efficiently inhibited photoaffinity labeling of HLA-A1, which is in agreement with the known HLA-A1 peptide-binding motif (acidic residue in P3 and C-terminal tyrosine). In contrast, photoaffinity labeling of HLA-A29 was efficiently inhibited by these as well as by the MAGE-3 and MAGE-6 nonapeptides. Finally, the HLA-B44 photoaffinity labeling, unlike the HLA-A1 and HLA-A29 labeling, was inhibited more efficiently by the corresponding MAGE decapeptides, which is consistent with the reported HLA-B44 peptide-binding motif (glutamic acid in P2, and C-terminal tyrosine or phenylalanine). The overlapping binding of homologous MAGE peptides by HLA-A1, A29, and B44 is based on different binding principles and may have implications for immunotherapy of MAGE-positive tumors.

Le mythe de la chute de Satan et la question du milieu d'origine de la Vie d'Adam et Ève

(Résumé de l'ouvrage) This collection of studies in honor of François Bovon highlights the rich diversity found within early expressions of Christianity as evidenced in ancient texts, traditions, symbols, and motifs. Old labels like 'apocrypha' or 'heresy' that for centuries have suppressed much of this evidence are removed, previous assumptions are questioned, and the old data are examined afresh along with the latest discoveries. The studies fall into six areas: ancient gospels, acts, early Christian movements, ancient interpretations, art, and manuscripts. Contributors include James Robinson, Helmut Koester, Harold Attridge, Karen King, and Jean-Daniel Kaestli.