Supervised PhD


Detection and classification of swallowing sound

Title

Preliminary study for detection and classification of swallowing sound

Candidate

Hajer Khlaifi

Abstract

The diseases affecting and altering the swallowing process are multi-faceted, affecting the patient’s quality of life and ability to perform well in society. The exact nature and severity of the pre/post-treatment changes depend on the location of the anomaly. Effective swallowing rehabilitation, clinically depends on the inclusion of a video-fluoroscopic evaluation of the patient’s swallowing in the post-treatment evaluation. There are other available means such as endoscopic optical fibre. The drawback of these evaluation approaches is that they are very invasive. However, these methods make it possible to observe the swallowing process and identify areas of dysfunction during the process with high accuracy. “Prevention is better than cure” is the fundamental principle of medicine in general. In this context, this thesis focuses on remote monitoring of patients and more specifically monitoring the functional evolution of the swallowing process of people at risk of dysphagia, whether at home or in medical institutions, using the minimum number of non-invasive sensors. This has motivated the monitoring of the swallowing process based on the capturing only the acoustic signature of the process and modeling the process as a sequence of acoustic events occuring within a specific time frame. The main problem of such acoustic signal processing is the automatic detection of the relevent sound signals, a crucial step in the automatic classification of sounds during food intake for automatic monitoring. The detection of relevant signal reduces the complexity of the subsequent analysis and characterisation of a particular swallowing process. The-state-of-the-art algorithms processing the detection of the swallowing sounds as distinguished from environmental noise were not sufficiently accurate. Hence, the idea occured of using an adaptive threshold on the signal resulting from wavelet decomposition. The issues related to the classification of sounds in general and swallowing sounds in particular are addressed in this work with a hierarchical analysis that aims to first identify the swallowing sound segments and then to decompose them into three characteristic sounds, consistent with the physiology of the process. The coupling between detection and classification is also addressed in this work. The real-time implementation of the detection algorithm has been carried out. However, clinical use of the classification is discussed with a plan for its staged deployment subject to normal processes of clinical approval.

Jury

  • Chairwoman: Sophie DABO, Professeur, University of Lille, Wooden bridge area, LEM UMR CNRS 9221 Laboratory
  • Referees:
    • Yannick KERGOSIEN, Professeur, University of Cergy Pontoise, UFR of Science and Technology
    • Mohamed Ali MAHJOUB , MCF-HDR, National Engineering School of Sousse, LATIS Laboratory
  • Examiners:
    • Monia TURKI, University Professor, National Engineering School of Tunis, University Campus - El Manar II
    • Sofiane BOUDAOUD, Maître de Conférences, University of Technology of Compiègne, BMBI UMR 7338
  • Guest:
    • Carla TARAMASCO TORO, Professeur, Escuela de Ingeniería Civil Informática Universidad de Valparaíso, Chile
  • Thesis Director:
  • Date and Location of the presentation: May 21st, 2019 at the University of Technology of Compiègne.

Papers

  • Hajer Khlaifi, Dan Istrate, Jacques Demongeot, Dhafer Malouche Swallowing Sound Recognition at Home Using GMM accepted for publication at Innovation and Research in BioMedical Engineering, June 2018

Statistical treatment of IMGT/HighV-QUEST NGS results.

Title

Statistical treatment of IMGT/HighV-QUEST NGS results for antigen receptors (immunoglobulins and T cell receptors): methodology and visualization.

Candidate

Safa Aouinti

Title

Abstract

This thesis is part of a project proposed by IMGTR , the international ImMunoGeneTics information systemR (http://www.imgt.org), created by Pr. Marie-Paule Lefranc in 1989 at Montpellier (Montpellier University, Laboratoire d’ImmunoGénétique Moléculaire (LIGM) à l’Institut de Génétique Humaine (IGH), UMR 9002 CNRS). The adaptive immune responses of humans and other jawed vertebrate species (gnathostomata) are characterized by the B and T cells and their specific antigen receptors, the immunoglobulins (IG) or antibodies and the T cell receptors (TR) (up to 2.1012 different IG and TR per individual). IMGTR , the international ImMunoGeneTics information systemR (http://www.imgt.org) built on IMGT-ONTOLOGY was created to manage this huge diversity. IMGT/HighV-QUEST is the first web portal, and so far the only one, for next generation sequencing (NGS) analysis of IG and TR big data sequences. One of its main features is the identification of IMGT clonotypes (AA) and in particular the analysis of their diversity and expression. This project has two major objectives. The first one is to generate a standardized statistical procedure for the analysis of IMGT/HighV-QUEST results. We evaluate the significance of the differences in the proportions of the diversity and the expression of the IMGT clonotypes (AA) between two batches of results per gene of a given group. In this object, multiple hypothesis tests are applied with adjustment of p-values via FWER control procedures (Bonferroni, Holm, Hochberg, SidákSS and SidákSD) and FDR (BH and BY). This methodology takes into account different levels of stringency based on biological studies which could be of great benefit and a breakthrough for comparative immunology. This approach has been implemented as a web application ‘IMGT/StatClonotype’ integrated into an R package ‘IMGTStatClonotype’ with corresponding visualization tools. In the context of NGS IG repertoires, the study of mutations is crucial. Somatic hypermutation is a normal process which, following the selection of mutated IG, leads to antibodies with increased affinity for the antigen. Mutational rate is a prognostic element used in clinical research for chronic lymphocytic leukemia. The IMGT/HighV-QUEST results comprise a detailed and standardized description of mutations. The second objective was to set up a statistical methodology to study somatic mutations in the V-REGION in order to study mutation patterns, to differentiate between selected mutations and to identify potential new alleles. We present a statistical approach based on a multinomial model for the analysis of somatic mutations specific to rearranged IG variable genes, as well as appropriate visualization methods. We use the IMGT unique numbering for a standardized description of mutations. An analysis tool in the form of a web application ‘IMGT/StatMutation’ integrated into an R package ‘IMGTStatMutation’ has been developed as a research support and will be made available to users with the corresponding visualization tools.

Jury

  • Chairman: Pr. Mohamed MNIF Professeur, École Nationale d’Ingénieurs de Tunis
  • Referee:
    • Pr. Ahmed REBAI (Professeur), Centre de Biotechnologie de Sfax
    • Pr. Nabil SEMMAR (Professeur), Institut Pasteur de Tunis, ISSBAT / UTM
  • Examiner : Dr.  Nahla ABDELLATIF (Maître de Conférences), École Nationale des Sciences de l’informatique, LAMSIN-ENIT
  • Thesis Director: Pr. Dhafer MALOUCHE (Professeur), École Supérieure de la Statistique et de l’Analyse de l’Information
  • Date and Location of the presentation: June 18th, 2019, ENIT, University Tunis-El Manar

Papers


Spatio-temporal modeling

Title

Spatio-temporal modeling of Zoonotic Cutaneous Leishmaniasis in Central Tunisia

Candidate

Khouloud Talmoudi Jabbari

Abstract

New statistical modeling techniques are developed to meet the need of various research fields. In this context, the analysis of temporal and spatio-temporal structured data is currently of great interest in epidemiological studies. The classical approaches, previously used, to model this type of data have been mainly based on linear assumptions. However, data related to meteorology, environment and epidemiology require the use of more flexible statistical models. In this work, we aimed to investigate the use of non-parametric and non-linear models to identify the significant factors associated with the incidence of a disease. Besides, we focused on the construction of a reliable predictive model able to predict outbreaks at spatial and temporal scales. The first contribution of the thesis was the use and examination of the non-parametric Generalized Additive Model (GAM) to explain the dynamic of the Zoonotic Cutaneous Leishmaniasis (ZCL) disease. This method allows a robust, smooth and flexible modeling of non-linear effects by means of splines. The smoothing splines are a set of functions which are able to model the effect of a covariate on a response variable without assuming any functional form of the relationship; instead the form is determined directly from the data. The model provided a visual tool to assess the partial effect of each explanatory variable on the response. It also enabled the inclusion of autoregressive error structures to perform a predictive model. This model was compared to the classical approach used, the Generalized Linear Model (GLM), in terms of performances and providing accurate predictions. The second contribution was the development of a nonparametric spatio-temporal Generalized Additive Mixed Model (GAMM) that estimates the space-time interaction in the case of ZCL disease. This model is carried out in three steps: First, we determine the error structure of the model. Next, we examine and compare different alternatives to account for space-time interaction. These alternatives consist mainly of the use of nonlinear random effect interaction between space and time and the use of a three-dimensional scale invariant tensor product, which is invariant to scales in time and space dimensions and constructed on the basis of penalized splines. Third, we identify the significant explanatory variables associated with the response variable, the number of ZCL cases. Finally, we developed a new statistical software as a third contribution which consists of a web application that we named ’EpiGAMapp’. This tool is implemented in the R programming language. It integrates interactive data visualization and offers the possibility to build a GAM model. It is useful for both health authorities and researchers who focus on predicting outbreaks on a disease based on the GAM approach

Jury:

  • Chairwoman: Prof.  Meriem Jaidane, Unité Signaux et Systèmes - Département TIC - Ecole Nationale d’Ingénieurs de Tunis - Université.

  • Reviewers:

  • Examiner: Prof.  Mohamed Ali El Aroui, University of Carthage. Tunisie.

  • Guest: Prof. Nissaf Ben Ben-Alaya, Université de Tunis el Manar.

  • Supervisors:

    • Prof.  Dhafer MALOUCHE, École Supérieure de la Statistique et de l’Analyse de l’Information.
    • Prof. Mohamed Chahed, Université de Tunis el Manar.

Papers

Sensory and consumer data mapping

Title

Statistical methods for sensory and consumer data mapping

Candidate

Ibtihel Rebhi

Abstract

Methodologies devoted to sensory data analysis are receiving increasing interest in recent decades by food industries as well as other activity sectors. The method is to gather ratings of human senses (e.g. taste, hearing, sight) in order to understand their perceptual behavior towards a set of products in the market. Two types of data are often collected : the consumer rating data that contains likings' scores and the sensory data which incorporates trained panelists' evaluations according to a list of sensory descriptors that characterize the perceived products. The background and purpose of this thesis are to gain insight into the connection between consumer and sensory data via External Preference Mapping technique. The output is a visualized map of optimal products and ideal characteristics, considered as a reliable tool by industries to make decisions about the development or the improvement of products. We assume that the obtained map suffers from a considerable lack of stability generated by high variability of human verdicts. This may lead to a mismatch between consumer requirements and products’ features; which may point to incorrect decisions and even product failure in the market. We propose therefore a series of best practices to enhance the stability of the sorting map. We introduce first a methodology to denoise consumer rating data using filtering methods. Denoising gives promising results; it allows to extract only the useful information about consumer likings which improves the mapping visualization. We produce a smoothed map and we find the optimal location of the product based on sensory data. Moreover, we tackle the issue of consumer preference prediction quality. Various models ranging from linear to non-linear regressions are advocated and compared using several information criteria. The model indicative of a good fit offers a more stable sensory space. As a way to compare sorting maps from different strategies, we have implemented an approach that provides an overall indicator of map’s stability. The method of analysis is based on a resampling process applied on consumer rating data associated with computations of predictions; distances recorded between sensory spaces. Thereafter, the efficiency of the proposed approach is evaluated by a simulation study and applications are illustrated on real data sets. Moreover, a major contribution of this thesis is to provide an easy-to-use tool that facilitates the adoption of sensory analysis techniques without requiring programming knowledge. We have developed a new software named SensMap implemented in R environment that includes a quick and efficient selection process for the different proposed features of mapping. The tool is associated with a friendly shiny interface named SensMapGUI. Both are made available on CRAN repository and can be easily used as a decision support tool devoted to researchers and industrial to understand consumers' behavior, meet their needs and maximize their liking. Applications illustrated on real data sets show that SensMap software provides constructive suggestions for product developers.

Jury

  • President: Mohamed MNIF from University of Tunis El Manar as President
  • Reviewer:
  • Examiner: Nabil Gmati from University of Tunis El Manar
  • Supervisor: Dhafer MALOUCHE from University of Carthage
  • Guest: Kaouther Ben Hassine from University of Carthage
  • Date and Location of the presentation: July 02, 2018, University of Tunis El Manar

Papers