08448380779 Call Girls In Diplomatic Enclave Women Seeking Men
Evaluating a row-store data model for full-content dicom management
1. Evaluating a row-store data model for
full-content DICOM management
Alexandre Savaris, Theo Härder, Aldo von Wangenheim
University of Kaiserslautern – Dept. of Computer Science – Kaiserslautern – Germany
Federal University of Paraná (UFPR) – Dept. of Informatics – Curitiba – PR – Brazil
National Institute for Digital Convergence (INCoD) – Florianópolis – SC – Brazil
5. DICOM content is:
• Structured at tag level
– Group/element ordered pair
– VR (Value Representation)
– VM (Value Multiplicity)
Modality
– (0008,0060)
– CS (Code String): 16 bytes maximum, accepting
uppercase characters, “0”-”9”, the SPACE character,
and underscore (“_”)
– 1 (a single value per tag)
Evaluating a row-store data model for full-content DICOM management 5 / 28
6. DICOM content is:
• Semi-structured at image level
– Tags are known at the evaluation (parsing) time
– The number/combination of tags varies according
to the data available at the examination time
– The number/combination of tags varies according
to the examination modality
– The number/combination of tags varies according
to the equipment manufacturer
Evaluating a row-store data model for full-content DICOM management 6 / 28
7. DICOM content is:
Evaluating a row-store data model for full-content DICOM management 7 / 28
Metadata + image
Metadata
Metadata
Metadata Patient
Study
Series
Image Image
Series
Image
Study
Series
Image
8. Storage in File Systems
Evaluating a row-store data model for full-content DICOM management 8 / 28
(0010,0020) PatientID
9. Storage in File Systems
Evaluating a row-store data model for full-content DICOM management 9 / 28
(0020,000D) StudyInstanceUID
10. Storage in File Systems
Evaluating a row-store data model for full-content DICOM management 10 / 28
(0020,000E) SeriesInstanceUID
11. Storage in File Systems
Evaluating a row-store data model for full-content DICOM management 11 / 28
(0008,0018) SOPInstanceUID
12. + Easy to organize and deploy
+ Easy to distribute over the network
+ Mounting points using NFS, for example
- Restrictive for query/retrieval
- Only the hierarchical level IDs are known without
file content evaluation
- Lack of indexes built over tag values
Storage in File Systems
Evaluating a row-store data model for full-content DICOM management 12 / 28
14. Storage in RDBMSs
Evaluating a row-store data model for full-content DICOM management 14 / 28
+ Easy to map the DICOM hierarchy into a set
of relations/relationships
+ Use of SQL for maintenance
+ Performance boost through indexes
- Need of a predefined DB schema
- Usually, composed by a restricted number of tags
- Scalability is “unnatural”
- Works well for single-node instances
- Multi-node instances are possible, but demand
considerable administrative efforts
16. What about NoSQL?
Evaluating a row-store data model for full-content DICOM management 16 / 28
• Native scalability
• Configurable partitioning/replication
• Loose constraints when compared to the
relational model (e.g., schemas, foreign
keys, referential integrity)
• Projected to work in the “huge” level
– Huge volumes of data, huge number of users, …
17. Two questions to be answered
1. Is it possible to manage full-content DICOM
images at tag level, using a data model built
over a row-store, NoSQL database?
2. Despite its close relationship with big
volumes of data, does a row-store, NoSQL
database perform well in scenarios of small
datasets when compared to known
approaches, i.e., relational databases?
Evaluating a row-store data model for full-content DICOM management 17 / 28
22. • Results include the time needed to parse/extract
individual tags from image files
• Storage time is derived from a combination of two
characteristics:
– The dataset size
– The file content complexity
• SA = 89.8% faster than CL (in cumulative query time)
– Communication and replication issues
• In CL, parallel writes are a solution to performance
improvement
– Speedup of 77.9% when compared to single writers
Results – Storage
Evaluating a row-store data model for full-content DICOM management 22 / 28
24. • Queries are executed by hierarchical level,
selecting values from tags related to each
level
• The row-store performs better when:
– There is high selectivity (the image level)
– The number of selected tags is minimal (the series
level)
• In general, RDBMS outperforms row-store
– 8.9% faster than SA
– 19.2% faster than CL
Results – Query
Evaluating a row-store data model for full-content DICOM management 24 / 28
26. • Retrieval operations are executed by
hierarchical level, returning sets of full-content
(metadata + pixel data) images
• Retrieval time decreases as selectivity increases
– CL setup for the row-store performs better than SA setup
– Partition by patientid contributes in routing retrieval
operations to single nodes
• In general, RDBMS outperforms row-store
– 81.7% faster than SA
– 83.2% faster than CL
Results – Retrieval
Evaluating a row-store data model for full-content DICOM management 26 / 28
27. Conclusions
1. Is it possible to manage full-content DICOM images at tag
level, using a data model built over a row-store, NoSQL
database?
– Yes. Row-stores are flexible enough to manage combinations of
DICOM tags in a consistent way.
2. Despite its close relationship with big volumes of data,
does a row-store, NoSQL database perform well in
scenarios of small datasets when compared to known
approaches, i.e., relational databases?
– According to the experiments performed in this work, no. The
row-store setups were outperformed by the RDBMS in the
overall evaluation.
– Other data models, however, can be better suited for the task.
Evaluating a row-store data model for full-content DICOM management 27 / 28