DICOM INTERNATIONAL CONFERENCE SEMINAR April 8 -10

DICOM INTERNATIONAL CONFERENCE & SEMINAR April 8 -10, 2008 Chengdu, China Medical Image Quality Assurance with Automated Constraint Validation Dongbai Guo Architect and Senior Development Manager Oracle Corporation db. guo@oracle. com, dbguo@ieee. org Nashua, NH 03062, USA

Market Trends • Growth in medical imaging – – New applications and new branches of medicine Increased utilization Increased spatial and temporal resolution Exponential growth in data volume • Computer aided detection (CAD) and diagnosis (CADx) • Data integration and centralization • Knowledge management, sharing and discovery

Impacts to DICOM • Wider, but sometimes hasty, adoption • Variant implementation quality • Non-compliant images lead to integration nightmare for – Clinical care – Research – Clinical trials • We need ways to specify and validate constraints

Validation Applications • Validate the conformance of a DICOM object with respect to – The DICOM standard (www. dvtk. org) – An institutional guideline – A jurisdictional requirement – A DICOM application

Formal Language Representation of Constraints • Formal language representation – Unambiguously specify constraints – Unambiguously exchange constraints – Object can be validated at the site of origination – Stand-alone validation services – Computer interpretable formal syntax

XML Encoding of Constraints • Use an XML document to specify predicates to evaluate against a DICOM object – Easy to exchange – Constraint can be managed, enhanced, version-controlled – Good readability – Wide range of tools – Using XML schema to validate syntax

Constraint Design

Sample Constraint, 1 (1) SOPCommon. Module: : = (1. 0) /* Based on P 3. 3 2007 table C. 12 -1 */ (1. 1) ( type 1(00080016) (1. 2) AND (1. 3) type 1(00080018) ) (1. 4) when false log (“SOP common module validation failed”); (1. 5) when false error (“Validation error: missing mandatory attribute for SOP common module”);

XML Mapping, 1 Based on P 3. 3 2007 table C. 12 -1 00080016 00080018 SOP common module validation failed missing mandatory attribute for SOP common module

Sample Constraint, 2 (2) CTImage : : = (2. 1) (2. 1. 1) ( (00080016==” 1. 2. 840. 10008. 5. 1. 4. 1. 1. 2”) (2. 1. 2) implies (2. 1. 3) (( type 1(0008) AND (0008 IN {“AXIAL”, “LOCALIZER”} ) ) (2. 2) AND (2. 2. 1) (00080016==” 1. 2. 840. 10008. 5. 1. 4. 1. 1. 2”) (2. 2. 2) implies (2. 2. 3) ( type 1(00280002) AND (00280002 ==’ 1’ )) (2. 3) AND SOPCommon. Module) (2. 4) when false warning(“Validation warning: not a valid CT image”);

00080016 1. 2. 840. 10008. 5. 1. 4. 1. 1. 2 0008 0008 < CODE_STRING>AXIAL LOCALIZER

Sample Constraint, 3 Context Sensitive Structure • The value of an attribute determines the structure of a DICOM object • Common in DICOM structured report – Example Table C. 17 -4 SR document content module attributes, PS 3. 3 – 2007, – Can be broken down to attribute level predicates (value(0040 A 040)==“TEXT” not. NULL(0040 A 160))

Validation Flow

Implementation • System implemented with PL/SQL, Java and C • Part of Oracle database 11 g. R 1 • Data model manages user-defined constraints • Changes to constraints does not interrupt database services • Can batch-validate DICOM content stored in or managed by a database

Quality Assurance (QA) • Validation of DICOM content – – By the capturing device At the PACS protocol layer Within an image archive At enterprise data exchange gateway • As a quality assurance service component of any service oriented architecture – Prevent application errors – Avoid costly rescan of patients – Eliminate waste in computation and network bandwidth

Using the QA Components • Distributed QA agent – Enable nation-wide radiation safety monitoring

Experiment Setup • A clinical trial requires images contain all DICOM image related tags for an image processing application – Verify part-10 file header exist – Verify all general image module tags exist – Image size (row/column) fell within the required range – Pixel data exist and is either not compressed or compressed with a compatible codec

Experiment Result • 15413 sample images collected from clinical trial sites • 121 of them failed validation. • All images passed validation can be analyzed by the image processing software

Conformance-aware Data Set • The same DICOM object can be verified by different QA requirements – Passed CT content validation – Failed CT CAD application validation • DICOM object and data flow can be associated with conformance results

Conformance-aware Dataflow

Summary • Formal language representation of DICOM constraint • XML encoding of constraint document • User-defined constraints • Conformance-aware data flow