Spot-Risc

chu_brest Partner’s presentation Purpose of partnership and context

LaTIM (Laboratory of Medical Information Treatment) gathers INSERM, Western Brittany University (UBO), Telecom Bretagne School and Brest Region University-affiliated Hospital in a close partnership to promote a research focused on therapeutic actions optimization through multimodal information integration. This challenge is divided into three main topics:

  • Physiological Information for musculoskeletal system;
  • Quantitative multi-modality imagery for diagnosis and therapy;
  • Indexation, traceability and integrity of multimedia medical information.

 An emerging axis of the laboratory aims at modelling relationships between biomechanical properties of bones and infectious processes evolution.

Joint infections (JIs), for several reasons, are raising issues in terms of statistical power.

An integration as well as a synthesis of JIs functioning appear to be of great importance and require to dispose of new tools of real-time and integrative analysis of infectious risks arising in surgery, constituting the challenge of this project entitled: complex bi and three-dimensional modellings of JIs based on multi-agents simulations.

  Solution developed by TEKLIKO
To materialize this project, TEKLIKO has developed in a first time a simple oriented platform prior to any research enabling to realize classical studies on heterogeneous data sets, in a complete transparency towards the data used and composed of:

  • A user web interface based on R (r-project.org) and Shiny (shiny.rstudio.com) disposing of :
    • A case-selection following criteria, from which random sampling;
    • Basic descriptive analysis and formulation of a summary;
    • A case for each unit : exploratory (clustering), analytical (regressions), causal approaches (Bayesian networks), predictive approaches (SVM, neural networks) ;
    • Personalised medicine: risk of adverse events, average delay of upcoming, (Bayes scenarii and SVM);
  • Integration interfaces of data derived from Brest Region University-affiliated Hospital units.

In a second time, this platform must have the ability to be enriched with less classical techniques of information treatment than usual statistical treatments and integrate multi-agents models.

Welcome page Continuous predictive analysis by Super Learner
Capture d’écran 2016-02-05 à 17.38.38 screenshot_spotrisc_superlearner_prostate
Bayesian network analysis Clustering by PAM/CLARA method
screenshot_spotrisc_superlearner_prostate_BN screenshot_spotrisc1