Éditeur(s) : HAL CCSD American Society for Microbiology
Résumé : International audience
Pneumocystis jirovecii pneumonia (PCP) is a life-threatening infection in immunocompromised patients. Quantitative real-time PCR (qPCR) is more sensitive than microscopic examination for the detection of P. jirovecii but also detects colonized patients. Hence, its positive predictive value (PPV) needs evaluation. In this 4-year prospective observational study, all immunocompromised patients with acute respiratory symptoms who were investigated for PCP were included, totaling 659 patients (814 bronchoalveolar lavage fluid samples). Patients with negative microscopy but positive qPCR were classified through medical chart review as having retained PCP, possible PCP, or colonization, and their clinical outcomes were compared to those of patients with microscopically proven PCP. Overall, 119 patients were included for analysis, of whom 35, 41, and 43 were classified as having retained PCP, possible PCP, and colonization, respectively. The 35 patients with retained PCP had clinical findings similar to those with microscopically proven PCP but lower fungal loads (P extless 0.001) and were mainly non-HIV-infected patients (P extless 0.05). Although the mean amplification threshold was higher in colonized patients, it was not possible to determine a discriminant qPCR cutoff. The PPV of qPCR in patients with negative microscopy were 29.4% and 63.8% when considering retained PCP and retained plus possible PCP, respectively. Patients with possible PCP had a higher mortality rate than patients with retained PCP or colonization (63% versus 3% and 16%, respectively); patients who died had not received co-trimoxazole. In conclusion, qPCR is a useful tool to diagnose PCP in non-HIV patients, and treatment might be better targeted through a multicomponent algorithm including both clinical/radiological parameters and qPCR results.
ISSN: 0095-1137

hal-01063906
https://hal.archives-ouvertes.fr/hal-01063906
DOI : 10.1128/JCM.01480-14
PUBMED : 25009050

Éditeur(s) : HAL CCSD
Résumé : International audience
We instrumented a realistic car simulator to extract low level data related to the driver's use of the vehicle controls. After proceeding these data, we generated features that were fed to a Multi-Layer Perceptron (MLP) and Support Vector Machines (SVM). Our goal was determine if the driver's Blood Alcohol Content (BAC) was over $0.4g.l^{-1}$ or not, and even estimate the BAC value. Our device process the vehicle's controls data and then outputs the user BAC. We discuss the results of the prototype using the MLP and SVM algorithms in both single-user and multi-user context for detection of drunk drivers and estimation of the BAC value. The prototype performed better with single user base than with multi-user, and provided comparable results with MLP and SVM.
ESANN 2013 proccedings,European Symposium on Artificial Neural Networks, Computational Intelligence and Machine Learning
Bruges, Belgium

hal-00850138
https://hal.archives-ouvertes.fr/hal-00850138

Éditeur(s) : HAL CCSD
Résumé : International audience
We propose using a distributed algorithm to count individuals in a population by way of a 3-phase algorithm. The first phase consists of detecting the presence of individuals, the second allows counting the number of individuals and the last aggregates the numbers of individuals over several paths. We propose an approach that uses a wireless sensor network built on a linear architecture. Here, each mote has a successor and a predecessor.
Wireless Communications in Unusual and Confined Areas (ICWCUCA), 2012 International Conference on

hal-00857695
https://hal.archives-ouvertes.fr/hal-00857695
DOI : 10.1109/ICWCUCA.2012.6402496

Éditeur(s) : HAL CCSD
Résumé : International audience
We propose using a distributed algorithm to count individuals in a population and to collect abiotic factors in order to characterize habitat. The first phase consists of detecting the presence of individuals from mapping ecotopes, the second allows counting the number of individuals and the last aggregates the numbers of individuals over several paths. We propose an approach that uses a wireless sensor network built on a linear architecture based on the concept of green tunnel. Here, each mote has a successor and a predecessor.
à venir

hal-00857694
https://hal.archives-ouvertes.fr/hal-00857694

Éditeur(s) : HAL CCSD
Résumé : International audience
Half away between edge detection and object recognition this study aim to characterize singularities in order to guide this two phases. The caracterization is obtained with a generalization of the Lipschitz exponents .for complex structures. This new characterization is called " maxima chain " and integrates the notion of value and spatial repartition of the maxima. The maxima chains are obtained with a multiscale decomposition using a Diadic Discrete Wavelet Transform (DDWT). We use the maxima of the gradient image computed with the wavelet details. The local maxima detect the position of irregular structures (edges) in the image (Canny algorithm). Then we characterize each maxima using their evolution across the different scales. A classification is made using the maxima chains. We integrate the classification to eliminate unwanted maxima (from noise or other objects) to drive the search of closed contours. In fact, there is a totally different signature from one maxima to another depending on the parameters of the objects : nature, shape, size, gray scale, texture, ...
RFIA Proceedings
Toulouse, France

hal-00602265
https://hal.archives-ouvertes.fr/hal-00602265