1 / 34

MAGIC-5

Distributed Computing Infrastructure (GRID). Computer Assisted Diagnosis (CAD). &. International Collaborations (CERN, CEADEN). HEP expertise on Image Analysis (CAD) - CALMA Grid Computing. Agreement with BRACCO. MAGIC-5.

remedy
Télécharger la présentation

MAGIC-5

An Image/Link below is provided (as is) to download presentation Download Policy: Content on the Website is provided to you AS IS for your information and personal use and may not be sold / licensed / shared on other websites without getting consent from its author. Content is provided to you AS IS for your information and personal use only. Download presentation by click this link. While downloading, if for some reason you are not able to download a presentation, the publisher may have deleted the file from their server. During download, if you can't get a presentation, the file might be deleted by the publisher.

E N D

Presentation Transcript

  1. Distributed Computing Infrastructure (GRID) Computer Assisted Diagnosis (CAD) & International Collaborations (CERN, CEADEN) • HEP expertise on • Image Analysis (CAD) - CALMA • Grid Computing Agreement with BRACCO MAGIC-5 INFN: Bari, Cagliari, Catania, Lecce, Napoli, Pisa, Torino Universities: Bari, Genova, Lecce, Napoli, Palermo, Piemonte Orientale, Pisa, Sassari Hospitals: Alessandria, Bari, Livorno, Milano, Napoli, Palermo, Pisa, Sassari, Torino, Udine Medical Applications on a Grid Infrastructure Connection Piergiorgio Cerello (cerello@to.infn.it)

  2. Sensitivity (positives/true positives) 73% - 88% Specificity (negatives/true negatives) 83% - 92% 2% - 10% increase with double reading CALMA • Breast Cancer Screening • Increased survival rate • Problems: costs and manpower • Computer Assisted Detection Piergiorgio Cerello (cerello@to.infn.it)

  3. Massive Lesions Microcalcifications SENSIBILITY: 92% SPECIFICITY: 92% SENSIBILITY: 94% SPECIFICITY: 95% Improved Sensitivity & Reduced Specificity Radiologist + CAD X + CALMA Radiologist + CAD X + CALMA A 82.8 (4.5) 94.3 (2.8) 94.3 (2.8) A 87.5 (3.0) 84.2 (3.3) 87.5 (3.0) B 80.0 (4.8) 88.2 (3.8) 90.0 (3.6) B 91.7 (2.6.) 85.9 (3.2) 88.4 (2.9) C 74.2 (4.0) 70.8 (4.2) 70.9 (4.1) C 71.5 (5.4) 82.9 (4.5) 87.1 (4.0) CALMA Results • Largest Database of digitised mammograms ( > 5000) • ROC (Receiver Operating Characteristic) Curve Piergiorgio Cerello (cerello@to.infn.it)

  4. Example: Italy 4 mammograms/exam (60 MB)/exam 6.7 Mpeople, 1 exam/2y 3.35 Mexams/year about 200 TB/year 1 PB/year on the European scale Huge amount of distributed data The “GRID philosophy” in mammographic CAD 2001: CALMA Open Issues • Virtually unlimited Database size • Intrinsically distributed Database – many sources • Network connections • Access required to all the images • Use Cases • Large Scale Screening • Teleradiology: diagnosis & training • CAD on demand Piergiorgio Cerello (cerello@to.infn.it)

  5. Accounting Storage Element “Blue” VO W Computing Element Data & Metadata Catalogues M S Monitoring “Green” VO Information System Authentication Authorisation User Interface Piergiorgio Cerello (cerello@to.infn.it)

  6. “Medical” (distributed application use case) Distributed databases, owned resources Special security needs: privacy Ease of installation, maintenance and access Small, single-purpose, single-VO dedicated grids An example: the GPCALMA project Medical Imaging communities Piergiorgio Cerello (cerello@to.infn.it)

  7. Data Collection Centre Diagnostic Centre Data & MetaData Catalogue Data Catalogue 1 - Data Collection 4 - Transfer Selected Data 3 - Run CAD remotely 5 - Interactive Diagnosis Data Catalogue 2 - Data Registration GPCALMA Screening CAD selection to minimize data transfers Piergiorgio Cerello (cerello@to.infn.it)

  8. GPCALMA Tele-training & Epidemiology Data Catalogue 3 - Spawn Processes 2 - Start CAD 4 - Remote Analysis 5 - Retrieve & Analyze Selected Images 1 - Data Selection Piergiorgio Cerello (cerello@to.infn.it)

  9. Storage Element 6 - Send CAD results Computing Element Computing Element 1 - Data Acquisition 2 - Data Registration 4a - Transfer Image 4b - spawn PROOF process Data Catalogue GPCALMA CAD on demand 5 - Run CAD algorithm 3 - Ask for CAD Piergiorgio Cerello (cerello@to.infn.it)

  10. PROOF ( http:// root.cern.ch ) AliEn ( http:// alien.cern.ch ) GPCALMA How to implement the above described Use Cases? • Move code rather than data • Share the images without moving them • Single VO in hospitals • Secure Access • Distributed Data Management • Scheduling of Computing Resources Piergiorgio Cerello (cerello@to.infn.it)

  11. The GPCALMA Graphic User Interface In use: Bari Napoli Pisa Sassari Torino Piergiorgio Cerello (cerello@to.infn.it)

  12. gpcalma.to.infn.it Server • Distributed System Configuration • Users’ Database • Data Catalogue • Web Portal Node Node Node Node Node • Client • Storage Element • File Transfer Daemon • ROOTd/PROOFd • GPCALMA Client Client Client Client Client The GPCALMA distributed system configuration • Clients installed: Lecce, Napoli, Pisa, Sassari, Torino Piergiorgio Cerello (cerello@to.infn.it)

  13. The AliEn-GPCALMA Core Serviceshttp://gpcalma.to.infn.it Piergiorgio Cerello (cerello@to.infn.it)

  14. Catalogue query Patient creation Image registration Piergiorgio Cerello (cerello@to.infn.it)

  15. 2 0 0 2 2 0 0 3 • CALMA algorithms rewritten in C++, based on ROOT • New GUI, with functionality to manipulate the images • AliEn server and clients operational • PROOF cluster configured • 1st mammogram remotely analysed in March 2003 • data/metadata structure being (re)defined • re-organisation of the CALMA Database • CALMA-DICOM format conversion The basic functionality is available and tested Demos presented at SC2003 and HG2004 GPCALMA Achievements Ongoing tasks: • C++ ROOT-AliEn API for Input Data Selection • improve the algorithms performance – new approaches • optimise the implementation of data and metadata • set up a prototype in the participating hospitals Piergiorgio Cerello (cerello@to.infn.it)

  16. GPCALMA CAD News • Masses • ROI Search • Features • AREA • Perimeter/AREA • Entropy • Fractal Dimension • Neural Network Piergiorgio Cerello (cerello@to.infn.it)

  17. 2.Masked Image 3. Obtained Binary Image 4. Connected Components Labelling Image 1.H-Dome Reconstructed Image GPCALMA CAD News • Microcalcifications Piergiorgio Cerello (cerello@to.infn.it)

  18. GPCALMA CAD News • Microcalcifications • Pre-Processing • Features • AREA • Perimeter/AREA • Neural Network • Classification: negative, benign, malignant Piergiorgio Cerello (cerello@to.infn.it)

  19. GPCALMA from GENIUS Piergiorgio Cerello (cerello@to.infn.it)

  20. GPCALMA on iBook Piergiorgio Cerello (cerello@to.infn.it)

  21. MAGIC-5 GPCALMA ANPI ADD COLON MAGIC-5 • INFN expertise and leadership in: • CAD development • Grid Middleware • Does any other Medical field but mammography require a similar approach? • CAD for Lung Cancer detection… it’s on time – like CALMA! • 3D CT images • search for different patterns • same Grid approach • AliEn is presently the best available Grid implementation in terms of easiness of installation, functionality, stability and scalability • Alzheimer’s disease diagnosis • Colonoscopy (?) • MAGIC-5 • 1 project (MAGIC-5) and common GRID Services • 3 Virtual Organisations • GPCALMA • ANPI (Analisi Neoplasie Polmonari in Italia) • ADD (Alzheimer’s Disease Diagnosis) Piergiorgio Cerello (cerello@to.infn.it)

  22. CAD for Lung Cancer? About 43 images/patient About 0.5 MB/image • 5 years survival rate for lung cancer: 14% (US), 10-15% (EU) • no improvement in the past 20 years • Low dose CT: 6 times more efficient than Chext X-Ray (CXR) in the detection of state I malignant nodules • CAD methods are being explored • Gurcan et al., Med. Phys. 29(11), Nov. 2002, 2552: “…computerized detection for lung nodules in helical CT images is promising…large variations in performance, indicating that the computer vision techniques in this area have not been fully developed. Continued effort will be required to bring the performances of these computerized detection systems to a level acceptable for clinical implementation.” Piergiorgio Cerello (cerello@to.infn.it)

  23. Best available trade-off between sensitivity for the detection of nodules and absorbed dose • Single(Multi)-slice: 1(1) tube + 1(N) detector array(s) with 500-900 elements + 1(4) DAQ channel: 1(2)D curved array, shorter scan time • N >= 4 detector arrays • (A)symmetric detector arrays • Detector elements or arrays can be combined to obtain different thickness and/or width • Collimators can also be used Spiral CT imaging principles • Linear patient motion through the gantry • Beam rotation • spiral pattern of data acquisition • one continuous set of volume data • Reconstruction options • (Slice reconstruction increment) • (Interpolation algorithm) • (Effective slice thickness) Piergiorgio Cerello (cerello@to.infn.it)

  24. N x P x S x T R Multi-slice vs. Single-slice • Volume Coverage: N= number of DAQ channels = 4 P= pitch (linear movement in T/beam collimation) S= detector width (mm) T= execution time (s) R= rotation time (s) = 0.5 s Piergiorgio Cerello (cerello@to.infn.it)

  25. 1 mm 120 KV 20 mAs Ric. 5 mm 120 KV 20 mAs Images: an example 5 mm 140 KV 120 mAs Piergiorgio Cerello (cerello@to.infn.it)

  26. Screening in Italy & EU-US • Main goal • reduce the death rate caused by lung cancer • The sample • 55-69y • >20 (packs/day) * y • Smokers (or ex-smokers < 10 y) • Agreement • No previous cancer • Italy • Ongoing programs: Genova, Milano, Torino • Starting phase: Regione Toscana – Emilia-Romagna • About 7000 exams in 4 years • EU – US • Collaborative Spiral CT-group • I-ELCAP: International Early Lung Cancer Action Project • EU ELCDG: EU Early Lung Cancer Detection Group • US: National Lung Screening Trial (50,000 people) Piergiorgio Cerello (cerello@to.infn.it)

  27. 1 SUBJECT 1 SUBJECT 150 SUBJECTS 15 SUBJECTS Minimal statistical value 10 MIN Best statistical value ? Neuroinformatics Portal • Interface for GRID applications • Statistical analysis of PET images databases for the study of the Alzheimer Disease • Alzheimer Disease (AD) is the leading cause of dementia, accounting for more than half of all dementias in elderly people • Why Grid? • Highly difficult collection of a control group built with normal images • Remote access to a database of normal patients • Access control (Cfr registration, autentication, certification) • Interactive SPM Statistical Analysis Piergiorgio Cerello (cerello@to.infn.it)

  28. Piergiorgio Cerello (cerello@to.infn.it)

  29. PORTAL The Alzheimer Diagnosis Use Case Univ. Ge, MiB, Osp. S. Raffaele SET of CONTROLS 1 (PET, SPECT IMAGES) SET of CONTROLS 2 (PET, SPECT IMAGES) STATISTICAL TOOL (SPM) UP LOAD SET of CONTROLS 3 (PET, SPECT IMAGES) IMAGE of PATHOLOGIC SUBJECT (PET or SPECT IMAGE) STATISTICAL ANALYSIS OF THE UPLOADED IMAGE SET of CONTROLS n (PET, SPECT IMAGES) Piergiorgio Cerello (cerello@to.infn.it)

  30. Portal AliEn Server SPM Server DB Catalogue PROOF Master AliEn Client SPM Server DB Reference Data collection Root Client AliEn Client SPM Server DB Reference Data collection Root Client AliEn Client SPM Server DB Reference Data collection Root Client Repository Node Repository Node Repository Node Server Alzheimer Disease Use Case Server Node User Node SPM Client Data Collection SPM Client Data Collection Piergiorgio Cerello (cerello@to.infn.it) User Node

  31. Alzheimer Disease Use Case User Node Image Acquisition Reference Atlas Selection Image Transfer Maps Visualisation 1 4 Repository Node Server Node Image Normalisation Data catalogue Query Image Transfer Statistical Analysis Maps Transfer Image Normalisation Image Comparison Results Transfer Repository Node Image Normalisation Image Comparison Results Transfer 2 2 3 3 Piergiorgio Cerello (cerello@to.infn.it)

  32. Piergiorgio Cerello (cerello@to.infn.it)

  33. Conclusions • Breast Cancer Detection in Screening Programs: good example of e-health application that would benefit from the use of GRID Services • The AliEn/PROOF based approach allows: • Minimisation of data transfers • Secure management of a distributed Virtual Organisation • The success will depend on: • the reliability and stability of interactive GRID Services • the performance of CAD algorithms: ongoing new approaches • the quality of the GUI • GPCALMA Virtual Organisation in the participating Hospitals • by the end of 2004 with improved CAD algorithms • New applications will follow • ANPI, ADD, COLON • EGEE/LCG/ARDA: Architecture Roadmap towards Distributed Analysis • Prototype developed in the framework of EGEE by Sep 2004 • Migrate to that prototype Piergiorgio Cerello (cerello@to.infn.it)

  34. Accounting Storage Element W Computing Element Data & Metadata Catalogues M S Monitoring Information System Authentication Authorisation User Interface “Blue” VO “Green” VO Piergiorgio Cerello (cerello@to.infn.it)

More Related