Analuse Globalisée des Données d ’Imagerie Radiologique Ecole Grid5000, Grenoble, 10 mars 2006 Recalage d'images médicales Validation expérimentale sur Grid5000 et EGEE Johan Montagnat, Tristan Glatard, Xavier Pennec Pierre-Yves Bondiau

Analyse Globalisée des Données d’Imagerie Radiologique Recalage d'images médicales AvantAprès

Analyse Globalisée des Données d’Imagerie Radiologique Ecole Grid5000, 10 mars 2006, Johan Montagnat (I3S) 3 Anatomically meaningful deformation Registration in 5 min on 15 PCs Adaptive non-stationary visco-elastic inter-subject registration Image Registration for Oncology

Analyse Globalisée des Données d’Imagerie Radiologique Ecole Grid5000, 10 mars 2006, Johan Montagnat (I3S) 4 Image Registration for Oncology Registration / segmentation are basic components of medical image analysis – Registration: finding homologous points / tranformation – Segmentation: give anatomical label to each image point Registration for brain radiotherapy – Planning  Fusion of image modalities (multimodal, rigid)  Warp atlas to patient image for segmentation (mono-modal, non-rigid)  Definition of Target volumes and Organs at risk: dose optimization – Follow-up (monomodal rigid) (ch 3/4)

Analyse Globalisée des Données d’Imagerie Radiologique Ecole Grid5000, 10 mars 2006, Johan Montagnat (I3S) 5 Image Registration for Oncology There is no universal registration algorithm – More than 600 references on medical image registration in 1997 – More than 100 papers each year… (70 at MICCAI 2004 only) Registration algorithms as Grid services – Use up to date algorithm – Evaluation / comparison of algorithm performances Challenges – Inter-operability (coordinate systems, transformation format…) – Ontology describing data, registration problems and algorithms

Analyse Globalisée des Données d’Imagerie Radiologique Ecole Grid5000, 10 mars 2006, Johan Montagnat (I3S) 6 Performance evaluation and validation Synthetic data (simulation): – Available ground truth – Difficult to identify and model all sources of variability Real data in a controlled environment (Phantom): – Possible gold standard – Performances evaluation in specific conditions  Difficult to test all clinical conditions  May hide a bias Image database representative of the clinical application – Usually no ground truth – Should span all sources of variability

Analyse Globalisée des Données d’Imagerie Radiologique Ecole Grid5000, 10 mars 2006, Johan Montagnat (I3S) 7 Bronze standard Bronze standard: The exact result is an unknown variable – Unbiased estimation: use redundant information  use many different registration algorithms (average biases, so that precision ~ accuracy)  Use many different data (redundant information to ensure precision)  Average transformations (maximal consistency) Best explanation of the observations (ML) : – Robust Fréchet mean – Robust initialisation and Newton gradient descent Result

Analyse Globalisée des Données d’Imagerie Radiologique Ecole Grid5000, 10 mars 2006, Johan Montagnat (I3S) 8 Multi-modality usecase

Analyse Globalisée des Données d’Imagerie Radiologique Ecole Grid5000, 10 mars 2006, Johan Montagnat (I3S) 9 Bronze standard workflow T, s

Analyse Globalisée des Données d’Imagerie Radiologique Ecole Grid5000, 10 mars 2006, Johan Montagnat (I3S) 10 MOTEUR workflow engine Development of a new execution engine : MOTEUR – compatible with Taverna workflow description language (Scufl) – Allowing data and component parallelism – Implementing result traceability Interfaces – Web Services – GridRPC (DIET middleware) Execution infrastructures > 2000 CPUs OAR batch submitter research infrastructure > CPUs, 5 PB LCG2 middleware (migration to gLite) production infrastructure

Analyse Globalisée des Données d’Imagerie Radiologique Ecole Grid5000, 10 mars 2006, Johan Montagnat (I3S) 11 workflow manager Service-based approach Grid User Interface Grid Resources Input 0 Service B Output 0 Input 0 Input 1 Service A Output 0 Data 0 Img Ref 0 Data 1 Img Ref 1 Data 2 Img Ref 2 Img Ref 1 Img Ref 2

Analyse Globalisée des Données d’Imagerie Radiologique Ecole Grid5000, 10 mars 2006, Johan Montagnat (I3S) 12 Test database acquired at the oncology department of Centre Antoine Lacassagne (Nice) MR T1 Images 256x256x120 voxels 16 bits/voxel (7.8 MB) 126 image pairs Test database

Analyse Globalisée des Données d’Imagerie Radiologique Ecole Grid5000, 10 mars 2006, Johan Montagnat (I3S) 13 EGEE vs Grid5000 Production grid (24/7 load) vs Experimental grid (resources volatility) EGEEGrid5000 Resources> 180 centers9 clusters > CPUs> 2000 CPUs > 5 PB~ TB Workload MgtResource BrokerGridOAR Batch SchedulerOAR (PBS, LFS...) Data transferGridFTPNFS

Analyse Globalisée des Données d’Imagerie Radiologique Ecole Grid5000, 10 mars 2006, Johan Montagnat (I3S) 14 Workfload Management System Short jobs (1 min), constant load (n jobs), measurements over 3 hours periods

Analyse Globalisée des Données d’Imagerie Radiologique Ecole Grid5000, 10 mars 2006, Johan Montagnat (I3S) 15 Data Management System Small number of data jobs (5), constant data size (7.8 MB), measurements over 3 hours periods

Analyse Globalisée des Données d’Imagerie Radiologique Ecole Grid5000, 10 mars 2006, Johan Montagnat (I3S) 16 Experimental results on the Bronze Standard application Critical size of Grid5000 not reached.

Analyse Globalisée des Données d’Imagerie Radiologique Ecole Grid5000, 10 mars 2006, Johan Montagnat (I3S) 17 Conclusions Grid modeling – Almost linear response time – Strong variability – Trade-off between grid pay-off and scale/achievable load MOTEUR inter-grid workflow manager – Load balance between different grids... –... should consider data transfer cost Real medical application for testing Data flow dominating medical applications – Need for a real data management service – Data manipulation cost to be taken into account – MOTEUR is relying on the grid data manager to avoid useless data transfer Interest in experimenting GridOAR for multi-cluster use of Grid5000