The research group ATLAS of GdR MADICS is organizing a workshop in Grenoble, on November, 22nd 2019. The objective of the workshop is to bring together fellow statisticians, computer scientists, signal and image processors, neuroscientists as well as doctors, biologists and people from the industry.

Registration to the workshop is closed as the maximum capacity of the seminar room, where the event will be hosted, is reached.

Organizers

  • Massih-Reza Amini, Professor, Laboratoire d'Informatique de Grenoble
  • Emilie Devijver, CRCN CNRS, Laboratoire d'Informatique de Grenoble
  • Charlotte Laclau, Associate Professor, Laboratoire Hubert Curien

Program

  • 9:15-9:45 Welcome Reception
  • 9:45-10:00 Brief Introduction of ATLAS
  • 10:00-10:45 Sophie Achard, DR CNRS, LJK
    Title : Assessing reliability of resting-state fMRI graph analysis: challenges in measuring brain connectivity networks alterations for clinical applications.
    Abstract : Resting state fMRI (rs-fMRI) datasets allow the observation of the functioning brain at rest. rs-fMRI combined with graph theoretical approaches has become popular with the perspective of finding network graph metrics as biomarkers in the context of clinical studies. The acquired data consist in multivariate time series. Each time series corresponds to the recording of a specific parcel of the brain for a finite duration. Each node of the brain graph is one time series and an edge in the network is characterizing an interaction between two brain regions. The identification of the graph edges is crucial and quantification of false edge detection needs advanced mathematical tools. In this talk I will present results on reliability of graph metrics in test-retest studies to assess the required sample size and acquisition parameters. The use of multiscale correlation with wavelets, multiple testing procedure and adequate graph metrics are advocated as an important pre-requisite for exploring clinical changes of brain networks. Different levels of analysis are derived and the choice of graph metrics is dependent on the reliability results. Clinical application on coma patients is presented. Alterations of brain networks are not found at the global level where global topological properties of complex brain networks may be homeostatically conserved under extremely different clinical conditions. However, in every patient we found evidence for a radical reorganization of high degree or highly efficient ``hub’’ nodes. Cortical regions that were hubs of healthy brain networks had typically become non-hubs of comatose brain networks and vice versa. Thus, Consciousness likely depends on the anatomical location of hub nodes in human brain networks.
  • 10:45-11:30 Luca Romeo, Postdoc Researcher, Università Politecnica delle Marche, Ancona, Italie
    Title: Designing and development of machine learning approaches for the early-stage prediction of different diseases using Electronic Health Record data
    Abstract: Recent years have witnessed an increasing amount of available Electronic Health Record (EHR) data and Machine Learning (ML) techniques have been considerably evolving. The temporal data stored in EHRs can be analyzed through ML algorithms to discover complex patterns and set-up powerful ML models that can be integrated into a Clinical Decision Support System to support the General Practitioners to screen patient population and predict the pathological risk condition. However, managing and modeling this amount of information may lead to several challenges such as overfitting, model interpretability, sparse observation over time and the natural unbalance of the predictive task. Starting from these motivations, in this talk I will present our work related to the design and application of sparse machine learning methodologies to increase the model interpretability, while implicitly managing high dimensional data and the usual unbalanced class distribution. Additionally, I will present a further work related to the introduction of a multiple instance learning-based algorithm for modeling the temporal evolution of longitudinal EHR data while dealing with small sample size and sparse observations (e.g., a small number of prescriptions for non-hospitalized patients). The proposed algorithms may represent the main core of a clinical decision support system.
  • 11:30-12:15 Nataliya Sokolovska, Associate Professor (HDR), Faculty of Medicine, Sorbonne Université
    Title: Learning interpretable cascade classifiers (applications in medicine)
    Abstract: In many prediction tasks such as medical diagnostics, sequential decisions are crucial to provide optimal individual treatment. Budget in real-life applications is always limited, and it can represent any limited resource such as time, money, or side effects of medications. The goal of cascade classifiers under budget constraints is to classify examples with low cost, and to minimise the number of expensive or time-consuming features (or measurements). I will present an approach to learn cost-sensitive heterogeneous cascading systems, I will briefly discuss other interpretable machine learning methods which we developed, and I will illustrate their efficiency by results on benchmarks and real medical tasks.
  • 12:30-14:00 Lunch break
  • 14:00-14:45 Cyril Boyault et Johann Poignant, CRCN CNRS, Institute for Advanced Biosciences.
    Title: Intelligence artificielle et réseaux de signalisation biologique - Conception de nouvelles options thérapeutiques en médecine
    Résumé: Les réseaux biologiques sont au cœur du fonctionnement du vivant et notamment des organes au sein desquels ils permettent la structuration de grands processus cellulaires comme la régénération ou le développement. Depuis quelques années, nous avons compris que des dysfonctions de ces réseaux participent au développement de nombreuses pathologies tels que les cancers, et ouvrent de nouvelles options thérapeutiques. A travers quelques exemples, nous présenterons les spécificités de réseaux de signalisation biologiques, avant de montrer les limites imposées par les connaissances actuelles. Nous discuterons des solutions proposées par des méthodes d'apprentissage de caractéristiques sur des graphes (Graph Convolutional Networks, node embeddings, ...) afin de surpasser ces problèmes.
  • 14:45-15:30 Michaël Sdika, Research Engineer, CREATIS
    Titre: Apprentissage profond pour l'imagerie médicale, une revue
    Abstract: L'apprentissage profond a changé la façon d'appréhender l'analyse d'image et le domaine médicale ne fait pas exception. Nous montrerons dans cette exposé comment il est utilisé pour résoudre des problèmes courant en traitement et analyse d'images médicales: segmentation, recalage, synthèse d'images, classification. Nous nous appuierons pour cela sur quelques résultats récents de l'état de l'art et nous présenterons également les limitations et tendances actuelles.
  • 15:30-16:15 Andrea Skanjeti, MD PhD, Hospices Civils de Lyon
    Title: Artificial intelligence in nuclear medicine
    Abstract: Lors de cette présentation on verra rapidement l’essentiel des procédures utilisées en médecine nucléaire, les défis actuels de cette discipline et la relation que cette discipline a maintenu avec l’intelligence artificielle dans le passé ainsi que au présent. Cependant, la conviction de l’auteur c’est que l’impact que l’intelligence artificielle aura dans le futur de la médecine nucléaire sera révolutionnaire, tout en nécessitant impérativement l’intervention humaine pour une application optimale. Très probablement l’intelligence artificielle ne remplacera pas les médecins nucléaires, les imageurs qui ne l’utiliseront pas seront remplacés par ceux qui profiteront des avantages de l’intelligence artificielle.

    Traduction depuis: https://www.deepl.com/fr/translator

    During this presentation we will quickly see the main procedures used in nuclear medicine, the current challenges of this discipline and the relationship that this discipline has maintained with artificial intelligence in the past as well as in the present. However, the author's conviction is that the impact that artificial intelligence will have in the future of nuclear medicine will be revolutionary, while imperatively requiring human intervention for optimal application. Most likely artificial intelligence will not replace nuclear physicians, imagers who do not use it will be replaced by those who enjoy the benefits of artificial intelligence.

Location

The event will take place in the IMAG building (salles de séminaires 1 et 2) located on Campus Saint-Martin d'Hères.

Address

Bâtiment IMAG - Université Grenoble Alpes
700 avenue Centrale - Domaine Universitaire
38401 St Martin d'Hères - France

Access by car

  • From Lyon or Valence, follow the direction Chambéry/ Rocade Sud. Take exit n°1 - Domaine Universitaire / Gières Mayencin.
  • From Chambéry, follow the direction Gières Domaine Universitaire. Take exit n°1 - Domaine Universitaire / Gières Mayencin.
  • From Grenoble centre town, take Avenue Gabriel Péri de St Martin d’Hères, then the entrance of the campus accross from Géant Casino.

Access by public transport

The site is easily accessible with two lines of the tramway : line B and line C. The tramway is adapted for use by people with reduced mobility. More informations on the public transportation network of Grenoble on the website of TAG or by phone, 04.38.70.38.70.
  • From SNCF train station, take the tramway Line B, Grenoble Presqu’île / Gières Plaine des Sports in direction of Gières Plaine des Sports. Tram stop: « Gabriel Fauré ».
  • Line C, Seyssins le Prisme / Saint Martin d’Hères Condillac Universités. Tram stop: « Gabriel Fauré ».