High-Performance Algorithms for Modeling, Simulation and Early Detection of Diseases in a Personalized Medicine Scenario

IDIFEDER/2018/032 (*)

This project focuses on the development of high-performance algorithms on graphic accelerator platforms and FPGAs for modelling, simulation and early detection of cardiological and related diseases, based on imaging techniques, physiological models and genomic data. These approaches are fundamental for a personalized medicine scenario in which specific characteristics of the patients are used on the models and classifiers. However, they require a non-trivial processing capacity.

The project has allowed to launch a high-performance computing platform based on graphic accelerators enabled to be offered as a production service. The equipment, after the adjudication of the competitions, has the following characteristics:

  • 14 processors Intel Skylake Gold 6130, with 14 cores each.
  • A total of 5,25 TB RAM.
  • 21 Graphic accelerators NVIDIA Tesla V100, with 32 GB RAM each.
  • 2 FGPAs Arria 10GX.
  • 1 Graphic accelerator AMD Instinct MI25.
  • 1 Graphic accelerator NVIDA Tesla P40.
  • 2 x 10GbE ports and 1 Infiniband emn port in each node.
  • Provided as a private cloud by means of OpenStack Rocky.

The equipment is currently being used to solve the following problems:

  • Deep learning models for the classification of noisy images between healthy, pathological and potentially pathological.
  • Storage and processing systems for large volumes of genomic data for knowledge generation through qualitative studies of genotype-phenotype association.
  • Clinical support tool based on the simulation of personalized cardiac models.
  • Development of numerical computation algorithms optimized for GPUs that simulate electrical propagation of the heart over long periods of time.
  • Development of tools to study the safety of drugs based on personalized cardiac models.
  • Computer tools to facilitate access to infrastructure resources in a private cloud format, using virtualization techniques and containers on resources that include specific hardware, such as GPUs.

In addition, it is allowing to establish collaborations with innovative companies of the Valencian Community in the same area:

  • Development of automatic diagnostic tools for rheumatoid heart disease on echocardiographic images, in collaboration with the company QUIBIM.
  • Reconstruction of images from the signal through full volume approximations, in collaboration with the company TESORO.

(*) Action co-funded by the European Union through the Operational Program of the European Regional Development Fund (ERDF) of the Valencian Community 2014-2020, under the objective: “Promover el desarrollo tecnológico, la innovación y una investigación de calidad”