using cloud & hpc infastructure to meet computing requirements for in-silico medicine

38
1 USING CLOUD & HPC INFRASTRUCTURES TO MEET COMPUTING REQUIREMENTS FOR IN- SILICO MEDICINE High Performance Computing & Big Data Conference 2016 Dr Susheel Varma, Chief Technology Officer, SSI Fellow Center for Computational Imaging & Simulation Technologies in Biomedicine – CISTIB The University of Sheffield, Sheffield, UK Susheel.varma@sheffield.ac.uk www.cistib.org

Upload: govnet-events

Post on 27-Jan-2017

161 views

Category:

Government & Nonprofit


2 download

TRANSCRIPT

From diagnostic imaging to image-based interventional planning of cerebral aneurysms

Using Cloud & HPC Infrastructures To Meet Computing Requirements for In-Silico MedicineHigh Performance Computing &Big Data Conference 2016Dr Susheel Varma, Chief Technology Officer, SSI FellowCenter for Computational Imaging & Simulation Technologies in Biomedicine CISTIBThe University of Sheffield, Sheffield, [email protected]

#

From Aerodynamics to Vascular Dynamics

2D Rotational Angiogram

Virtual FFR using Computational Fluid Dynamics

#

#

#

CellTissueOrganismOrgan Proteins

Complex Spatio-temporal Modelling

#

#

#

Computational Imaging & PhysiologyDescriptive & predictive computational models of physiology and post-interventional disease courseNon invasive and in vivo visualization of biological structure & functionComputerized high-throughput quantification of structure & function from images and their fusionCISTIBBiomedicalImaging & SensingImage & SignalComputingPersonalisedModeling & SimulationSubject-specific biomedical simulationsVirtual deployment of medical devicesTraining systems for minimally invasive interventionsSubject-specific design or customization of medical devicesPre-operative interventional planningImage-based surgical and interventional guidanceEvaluation of targeted contrasts agentsVirtual imaging techniquesStructure & function quantification from medical imagesVisualization & fusion of multimodal imagingAdvanced diagnostic and prognostic imaging biomarkersQuantify impact of medical products on structure & function

CardiovasculareuHeartCerebrovascular@neurISTMusculoskeletalMySpine / MD-PaedigreeNeurodegenerationVPH-DARE@IT

#

in silico Medicine (Precision Medicine)

is the direct use of computer simulation in the diagnosis, treatment, or prevention of a disease.Predict diseasePersonalise treatment

#

10 INSIGNEO 2015

IMAGES/DATA

PROCESS

ANALYSIS

I.T.

CLINI CAL

OUTPUT10% RUPTURE RISK

PORTAL

NETWORKS

WORKFLOWS

CLOUD/HPC

In silico workflows are built using two parallel strands:In Silico workflows

#

IMAGES/DATA

PROCESS

ANALYSIS

OUTPUT

NEUROLOGY (e.g. @neurIST)MRI SCANSEGMENTATIONRUPTURE RISK

IMAGES/DATA

PROCESS

ANALYSIS

OUTPUTBLOOD FLOW

CARDIOLOGY (e.g. euHeart)ANGIOGRAMSEGMENTATIONPRESSURE PROFILEFLUID DYNAMICS

IMAGES/DATA

PROCESS

ANALYSIS

OUTPUT

ORTHOPAEDICS (e.g MySpine)MR/CT SCANSEGMENTATIONDISC DEGENERATIONBENDING

I.T.

NETWORKS

CLOUD/HPC

WORKFLOWS

PORTAL

In Silico workflows for every medical domain

#

Clinical Research Exemplars

#

euHeart: Patient-Specific Cardiac Simulation Workflow

#

euHeart: Patient-Specific Cardiac Simulation Workflow

#

Image-based Computational Haemodynamics

DICOMInput: DICOMOutput: 3D imageDescription: Converts a DICOM image to VTK image

Volume RenderingInput: 3D imageOutput: 3D imageDescription: aneurysm and vessels Visualisation

Bounding BoxInput: 3D imageOutput: ROIDescription: volume selection

GAR SegmentationInput: Image,ROIOutput: surface meshDescription: vessels and aneurysm extraction

Mesh EditingInput: surface meshOutput: surface meshDescription: clipping vessels, cleaning surface (cell removal, closing holes, smoothing)

SkeletonizationInput: surface mesh.Output: skeleton.Description: necessary to set the boundary conditions

Aneurysm isolationInput: surface meshOutput: surface meshDescription: aneurysm isolation

Morphology DescriptorsInput: surface meshOutput: xml, vtk Description: surface, depth and ZMI calculation

Volumetric MeshInput: surface meshOutput: volumetric meshDescription: creates a volumetric mesh of the selected geometry

Flow SimulationInput: volumetric mesh, cclOutput: wall shear stress mapDescription: solves flow equations

Flow Simulation post-processingInput: wall shear stressOutput: .csv fileDescription: computes hemodynamic descriptors

CFD preprocessor

Input: xml, surface meshOutput: surface mesh, cclDescription: Defines hemodynamic model

Input: surface, 1D modelOutput: xml, vtkDescription: boundary conditions for CFDSelecting Boundary ConditionsNeck Selection

Input: surface meshOutput: surface meshDescription: aneurysm neck surface and dome selection

GIMIAS@neuFuseANSYS (ICEM)ANSYS (CFX)Manual interactionCommon operationsMorphological analysisHemodynamic analysis

#

Image-based Computational Haemodynamics

Image acquisitionSegmentationSurface correctionVolumetric mesh generationComputational Fluid Dynamics modeling and simulationData analyses

Patient specificboundary conditions

OSIWSSStreamlines

#

VPH-Dare@ITDementia

#

Model of brain life course and ageing

World Alzheimer Report 2014 www.alz.co.uk/research

#

Hypothetical model of biomarkers in AD

Jack CR Jr, Knopman DS, Jagust WJ, Shaw LM, Aisen PS, Weiner MW, Petersen RC, Trojanowski JQ. Hypotheticalmodelofdynamicbiomarkersof theAlzheimer'spathologicalcascade. Lancet Neurol. 2010 Jan;9(1):119-28.Petersen RC. Alzheimer's disease: progress in prediction. Lancet Neurol. 2010 Jan;9(1):4-5. Jack CR Jr1, Knopman DS, Jagust WJ, Petersen RC, Weiner MW, Aisen PS, Shaw LM, Vemuri P, Wiste HJ, Weigand SD, Lesnick TG, Pankratz VS, Donohue MC, Trojanowski JQ. Tracking pathophysiological processes in Alzheimer's disease: an updated hypothetical model of dynamic biomarkers. Lancet Neurol. 2013 Feb;12(2):207-16.

#

Population Data/Imaging

500k people100k imaged

#

20

Biophysical Brain ModelInterpolated permeability tensors: Input to MPET

Mesh Generation Workflow

Philips Brain Segmenation

Diffusion Tensor Extraction

#

Biophysical Brain Model

Vardakis JC, Tully BJ, Ventikos Y. Exploring the efficacy of endoscopic ventriculostomy for hydrocephalus treatment via a multicompartmental poroelastic model of CSF transport: a computational perspective. PLoS One. 2013 31;8(12):e84577.

Source: http://www.3dscience.com/Individual / Population ProfilesAnatomical ProfileTissue Types &PropertiesGeneticProfile

Environment / Lifestyle ProfilesSystemic BCAlterationsMolecular AlterationsGeneticAlterations

CellularProfile

#

Environmental & Lifestyle Factors

Source: Environmental Threats to Healthy Aging. http://www.agehealthy.org/

#

VPH-DARE@IT in a nutshellSection name

#

Multiscale Multifactorial Multiparadigm Modelling Platform

Human health dataclinicalpopulationenvironmental

Mechanistic ModellingTop-down

Phenomenological ModellingBottom-upStatistical associationsConnectivity networksBiophysical FE ModelsMetabolic PathwaysBiosignals Modelling

personalised environment &informationpersonal & environmentaldata

Platforms for Biomedical Research &Clinical Decision Supporthealthcareresearchcitizensmechanisticpredictions &biomarkersphenomenological inferences & associations

#

VPH-DARE@IT Partners

The University of Sheffield VTT Technical Research Centre of FinlandESI Group S.A Advanced Simulation & Design GmbH Empirica Gesellschaft fr Kommunikations und Technologieforschung mbH Universitetet i Oslo Erasmus Universitair Medisch Centrum RotterdamHirslanden Klinik Philips Medical Systems Nederland BV Eidgenssische Technische Hochschule ZrichKings College, London Philips Technologie GmbH Sheffield Teaching Hospital NHS Foundation Trust University College London It-Suomen yliopisto University of Maastricht Kinematix (Tomorrow Options Microelectronics S.A.)Imperial College of Science, Technology and Medicine EIBIR Gemeinntzige Gmbh zur Frderung der Erforschung der Biomedizinischen Bildgebung

#

VPH-ShareResearch As A Service

#

INSIGNEO 2015

Cloud Platform(Public / Private)

Patient DataWorkflow InputsWorkflow Outputs

Semantic Services

Patient Centred In Silico Workflows

Patient AvatarApplicationsInfrastructureHPC Infrastructure(DEISA / PRACE)Personalised ModelKnowledge DiscoveryData InferenceCompute ServicesStorage ServicesKnowledge ManagementData Services:Patient/PopulationeuHeart@neurISTVPH OPViroLab

Select WorkflowRetrieve Existing DataTransform or Infer DataRun WorkflowReturn Results

Outreach

#

Single point of entryApplicationRepositoryRich library of biomedical dataCloud PlatformVPH ApplicationsBuild new workflowsHigh Performance ComputingGuided search

#

Single point of entryApplicationRepositoryRich library of biomedical dataCloud PlatformVPH ApplicationsBuild new workflowsHigh Performance ComputingGuided search

#

VPH-Share Technology Architecture

ATMAtmosphere Cloud Platform

Atomic Service Deployment Wizard

MAFEventBus

Authentication Services

Workflow Execution Service

Workflow Registry

Atomic ServiceRegistry

Atomic ServiceManagerData Browser

Atomic ServiceGeneric InvokerMaster Interface

Cloud Facade

Visualisation Tools

Workflow Composer

VPH-Share Client

Generic WorkflowDocument

Atomic Service Description

CloudClients

libcloud provider

libcloud provider

Monitoring Controller

High PerformanceExecution Engine (AHE)

Extension Points

SPRUCE

HARC

Steering

AHE Services API

AHE Runtime

AHE Engine

App Registry

JBPM Workflow& Main Logic

AHE DatabaseHibernate ORM

App State Object

Storage Module

Connector Module

External Data Storage

External HPC PlatformSecurity Module

Allocation Management Service

AMS Manager (Java) OSGi bundlesApache Karaf

Scheduler / Optimizer

Algo n

Algo 1

REST API & HTML Service(Ruby) Sinatra & Passenger

Domain Model (Ruby)

Atmosphere Internal RegistryMongoDBVirtual Machine Template Registry

Data Buckets(C-DISC, CSV, )Databases(SQLServer, )

External StructuredData Providers

DataPublishingSuite (GUI)

Schema Crawler

SPARQL

Discovery

Browser

Search

RDB2RDFService

LOD Databases

Silk, LinQuerService

LD DatabasesMulti-Ontology/Archetype Search Services

Taverna Server

Service Registry

Load Balancer

Proxy Controller

Data Reliability & Integrity Services

PSLoader

External Cloud Data StorageSemantic Services

VoID Document Database

Database 1 Query Services(SPARQL & SQL)

Database Services Integration PointsDatabase 2 Query Services(SPARQL & SQL)

Database n Query Services(SPARQL & SQL)

Individual Relational Databases

VoID Services

VoIDDocument

Atmosphere Cloud Platform

Monitoring System

Atomic Service Instance Contents

Raw Operating System (Linux)

LOBCDER Federated Storage AccessRoot Volume

VPH-Share Tool / App

Web Service WrapperSoaplab2, CXF, soap4r

Remote AccessService

Web Service Security Agent

Monitoring Agent (Munin)

Hypervisor

Driver

Manager

Compute Worker

Network WorkerObjectStorage(Swift)

Dashboard

Queue

Scheduler

ProxyAccountContainerObject

ASIProxy

Private Compute & Storage Cloud (OpenStack Example)

Data Volume

Data Resource Catalog

LOBCDER Data Federation Middleware

Data Stores

Connection Module

Request Manager

Access & Control Frontend

Virtual Resource System

Driver 1

Driver n

Cloud Storage Driver

Data Infrastructure Services

Images

NOVA API

#

32#VPH2014 Trondhiem 09-Sep-14

Physicalresources

Atomic Service InstancesDeployed by AMS on available resources as required by WF mgmt or generic AS invoker

Raw OS (Linux variant)

LOB Federated storage access

Web Service cmd. wrapper

Generic VNC server

VPH-Share Tool / App.

DRIService

Atmosphere persistence layer (internal registry)

VM templates

AS images

Available cloudinfrastructureManageddatasets

101101011010111011

101101011010111011

101101011010111011

AMService

LOB federatedstorage access

Cloud stackclients

HPC resourceclient/backend

Data and Compute Cloud Platform

VPH-Share Master UI

AS mgmt. interfaceGeneric AS invokerComputationUI extensions

Data mgmt. interfaceGeneric data retrievalData mgmt.UI extensions

Remote access toAtomic Svc. UIs

Custom AS client

Workflow description and executionDeveloper

ScientistAdmin

Security mgmt. interfaceSecurityframework

Web Service security agent

Modules available in first prototype

Cloud/HPC Platform Architecture

#

Tabular DataNon-Tabular DataClinical Information Systems

Data Publishing SuiteSemantic Services

Computational Workflows and Services1234891067

MedicationsVital SignsLab ReportsDemographicVital SignsImagesDemographicRisk FactorsGenomic DataParameterEstimationUncertaintyPropagationPatient Avatar

RDFGraphs

ReferenceData

PhysiologicalEnvelope

#

ClinicalResearcherWorkflow Manager API

VPH-Share pluginTavernaServer

VPH-Share WorkflowCloudFaade

Web-basedRemote Desktop

AS without interactionAS with interactionCLIENT-SIDESERVER-SIDEASASASASASASExternalApplication

STORAGE

VPH-Share pluginTavernaWorkbench

Web servicesGIMIAS CLPs

VPH-Share pluginTavernaOn-lineWeb services

#

VPH-Share Platform by Numbers

#

VPH-Share Platform by Numbers4 Private Data CentersCYFRONET, KrakowUoS, SheffieldSTH, SheffieldUNV, Vienna800vCPU Cores, 32TB RAM, 200+ Applications (Baseline), 453 Applications (Peak)250TB+ Structured/Unstructured Data Storage150+ Scientific Applications [VMs -> Docker]50+ Scientific WorkflowsPublic Cloud Burst (Avg 2k CPUhrs/month)

#

36

VPH-Share Platform by NumbersA variety of projects are already making use of VPH-Shares infrastructure services for running workflows and storing tools and data.

#

ChallengesDistributed data lifecycle management is really hard!!Use the right level of metadata to avoid a collective prisoners dilemmaApply provenance metadata at all stages of the data pipelineInter- and intra-operability between clouds and HPCCommunity level push towards standardised access protocolDealing with batch & unbounded data streams (lifestyle)

We dont want to be in the business of build custom infrastructureWe want to be in the business of doing Science

#