Radiomics Data Management, Computation, and Analysis for QIN F2F 2016
Download as PPTX, PDF2 likes532 views
Large Scale Data Management Computation and Analysis for Quantitative Imaging Research Talk at the 2016 QIN Annual Meeting — covers resources developed for the Quantitative Imaging Network. Includes TCIA data curation, APIs, supported data types, as well as co-located computing and systematic phenotyping of imaging biomarkers
Ad
More Related Content
What's hot (20)
Viewers also liked (19)
Ad
Similar to Radiomics Data Management, Computation, and Analysis for QIN F2F 2016 (20)
Ad
Recently uploaded (20)
Radiomics Data Management, Computation, and Analysis for QIN F2F 2016
- 1. 1 Large Scale Data Management, Computation, and Analysis for Quantitative Imaging Research Ashish Sharma, Ph.D. Department of Biomedical Informatics, Emory University Fred Prior, Ph.D. Department of Biomedical Informatics, Univ. of Arkansas for Medical Sciences
- 2. 2 High Quality Data The Role of Data Curation Data Dump or Information Resource
- 3. 3 Challenges to Advances in Quantitative Imaging “Data! Data! Data!” he cried impatiently. “I can’t make bricks without clay.” — Sherlock Holmes • Large, well-curated, repositories • High Quality Data • Radiology, Pathology, RT • Clinical, Genomic • Derived Feature Sets • Security and Access Control • Flexible Access • Co-Located Computation • Reproducible pipelines • Systematic, Reproducible tools for imaging phenotype/biomarker exploration
- 4. 4 Resources Provided by the UAMS + Emory QIN Features
- 5. 5 Built on The Cancer Imaging Archive TCIA encourages and supports cancer- related open science communities by hosting and managing the data archive,, and relevant resources to facilitate collaborative research.
- 6. 6 RT Collections and Curation Process Partial collections: • RTOG 0522 • NRG-1106 (private) • NRG-1308 (Private) TCGA-HNSC Additional collections in process or planned
- 7. 7 Automated RT, Image Set Curation Individual Bulk Correction Reduce the time needed to correct data inconsistencies Curation Time: Days few minutes (15 Collections 120K images) Extend POSDA to all curation related checks and activities
- 8. 8 Digital Pathology caMicroscope • Visualization of Pathology Images and Features • Data Management • Native Support for Scanner Format (Scanner caMic) • Dynamic Services (launch analysis from browser) • Scientific Mashups (DataScope and DataCafe) Supported by NCIP/CTIIP Project; ITCR U24
- 9. 9 Pragmatic Randomized Trial of Proton versus Photon Therapy for Locally Advanced Breast Cancer • Treatment plans for Photon (left) and Proton (right) Adjuvant therapy for Lung cancer • Dose Value Histogram Demonstrates that with IMRT, 50% of the heart receives ~ 36 Gy, while proton delivers < 3Gy. Dose Value Histogram Intensity Modulated Radiation Therapy Proton Therapy
- 10. 10 Data Access REST APIs to access TCIA Data APIs to access protected data (coming soon) Mashups from fusion of diverse datasets to support • Data exploration • Hypothesis formulation • Discovery of latent relationships and patterns
- 11. 11 Reproducible Science Incentives for data sharing TCIA assigns DOIs for individual collections as well as data subsets that were used in publications One click access to data via DOI Use this DOI to get a Nature Scientific Data publication Will be developing mechanisms to track usage
- 13. 13 Radiomics Portal (Galaxy + Lessons from Connectome Workbench) Fast access to co-located data Imaging, RT, Radiomic Features (TCIA) Clinical and other structured (i2b2/Eureka!) Lung Segmentation Pilot Tool Dissemination
- 14. 14 Lung Analysis Pipeline 1. Segment lung field from CT 2. Use three parallel segmentation algorithms to identify all closed structures 3. Voting algorithm produces final segmented regions 4. Feature Extraction (shape, texture, margin,...) 5. Feature Classification (optional) 6. Image-based phenotype Top, Green Expert Manual Segmentation Bottom, Blue Automated Segmentation The results of three independent automatic segmentation algorithms are combined using a voting mechanism to arrive at final automatic segmentation.
- 15. 15 Biomarker Exploration Pipeline • Data Loading: Radiomics data is integrated into Eureka! Custom sources can be incorporated via REST APIs • Construct Radiomic Phenotypes: Using the Eureka GUI, users can explore their radiomics data and construct radiomic phenotypes, that incorporate clinical, imaging and computed features. These phenotypes are captured as graphs and stored in a graph database (Neo4J) i2b2 TCIA MongoDB & Others… Radiomic Phenotypes
- 16. 16 Eureka! Clinical Analytics • Open source (Apache 2 license) RESTful clinical phenotyping system Clinical phenotyping: computing temporal patterns from EHR data reflecting disease, therapeutic response and prognosis • Supports loading data and computed phenotypes from data warehouses and flat files into i2b2, Neo4j and flat files • One REST service + web interface controlling many ETL processes setup with configurable metadata stored in an extended i2b2 metadata schema
- 17. 17 Use Cases • Construct phenotypes by fusing clinical and derived feature data • Compute phenotypes on available data • Ongoing pilot with NLST data • LungRADS • Study changes to LungRADS screening recommendations by incorporating demographics
- 18. 18 Summary • TCIA is a stable yet continually evolving and expanding information resource for QIN and other Cancer Imaging initiatives • Radiomic & Radiogenomic analyses require information fusion and HPC pipelines to deal with large subject populations and advanced computationally intensive algorithms • We are exploring the potential of OODA graph statistics for multi-parametric biomarker discovery and validation.
- 19. • NCI QIN: Resources for development and validation of Radiomic Analyses & Adaptive Therapy — PI: Prior, Sharma (UAMS, Emory) • NCIP/Leidos caMicroscope — A Digital Pathology Integrative Query System; PI: Sharma • PCORI, Pragmatic Randomized Trial of Proton versus Photon Therapy for Locally Advanced Breast Cancer to compare the effectiveness of photon vs. proton Radiotherapy. (PI: Bekelman) • Leidos Biomedical Research, Contract 16X011 for NCI, Maintenance and Extension of The Cancer Imaging Archive (TCIA), (PI: Prior)
Editor's Notes
- #2: T