html5-img
1 / 29

Determining Scanned Body Part from DICOM Study Description for Relevant Prior Study Matching

Determining Scanned Body Part from DICOM Study Description for Relevant Prior Study Matching. Thusitha Mabotuwana , Yuechen Qian Philips Research North America 21 August 2013. Typical radiology workflow. Imaging order (fax/e-referral). HL7. Exam Protocoling. HL7. HL7.

rusty
Download Presentation

Determining Scanned Body Part from DICOM Study Description for Relevant Prior Study Matching

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. Determining Scanned Body Part from DICOM Study Description for Relevant Prior Study Matching Thusitha Mabotuwana, Yuechen Qian Philips Research North America 21 August 2013

  2. Typical radiology workflow Imaging order (fax/e-referral) HL7 Exam Protocoling HL7 HL7

  3. Typical radiology workflow Imaging order (fax/e-referral) DICOM HL7 Reporting Exam Protocoling DICOM HL7 HL7

  4. Typical radiology workflow Imaging order (fax/e-referral) DICOM HL7 Reporting Exam Protocoling DICOM HL7 HL7

  5. Background • Radiologists need to understand the clinical context when reading a new study. • Most relevant prior study is often used as the reference to compare current findings against.

  6. Current status • Important to find the correct prior study for comparison. • Determining relevant prior study is not always straight-forward, especially with complex patients having many (e.g., >20) studies for multiple conditions.

  7. Current status – opportunity for improvement • Typically, matching is done based on scanned body part (e.g., Head, Abdomen) corresponding to Body Part Examined field in DICOM (0018, 0015) • Body part field of the DICOM header is fairly generic • e.g., Study done to exclude pancreatitis and another study done to exclude renal stones will both have their body part field set to “abdomen”

  8. A typical DICOM header

  9. Our approach • Identified other DICOM attributes containing anatomy related information. • DICOM Study Description (0008, 1030) field • e.g., CT CHEST ABD/PEL LIVER • DICOM Protocol Name (0018, 1030) field • e.g., C/A/P W/ARTERIAL LIVER/Abdomen • DICOM Series Description(0008, 103e) field • e.g., LUNGS, Coronal

  10. DICOM Study Description field • Narrative, free-text and institution-specific terms (i.e., non-standardized). • Abbreviations (e.g., UE – upper extremity) • Synonyms (e.g., neuro) • Procedure names (e.g., mammogram) • Can an algorithm be developed to reliably extract the most specific anatomy information from DICOM Study Description field?

  11. System overview Preprocessing (e.g., stop word removal) Word normalization (e.g., global abbreviation replacement – ‘LT’) Procedure-related anatomy extraction (e.g., mammogram) Anatomy extraction using word combinations Postprocessing of extracted anatomies

  12. Algorithm development – reference taxonomy - 160 unique concepts were included in the taxonomy with an additional 87 terms included as synonyms or abbreviations

  13. Algorithm development – reference taxonomy features • Parent-child relationship between concepts. • Multiple abbreviations and/or synonyms. • Child concepts inherit properties from parents(e.g., laterality) • Concepts should be interpreted within the context of its ancestor (e.g., ‘soft tissues’ may appear multiple times in the taxonomy, but MRI P FACE SOFT TISSUE W would match to ‘face soft tissues’). • The taxonomy is a representation of intention of procedure (e.g., a mammogram study description – MAM BILAT DIGITAL W/CAD does not explicitly mention ‘breast’).

  14. Input string MRI ANKLE/FOOT WWO LT Empty or null string Return Remove stop words and special characters; convert to lower case mri ankle foot lt Replace known global abbreviations (e.g., LT with left) mri ankle foot left Determine modality and replace ankle foot left i) ankle foot left ii) ankle foot foot left ankle left Determine word combinations for input string in descending order Yes left foot left ankle Yes Intelligent filtering Show children concepts? No Empty input string or all combinations processed left foot No Process next word combination Yes Add matched term to list Match found? No Yes Replace matched terms from input string All combinations for current iteration finished? No

  15. Input string MRI ANKLE/FOOT WWO LT Empty or null string Return Remove stop words and special characters; convert to lower case mri ankle foot lt Replace known global abbreviations (e.g., LT with left) mri ankle foot left Determine modality and replace ankle foot left i) ankle foot left ii) ankle foot foot left ankle left Determine word combinations for input string in descending order Yes left foot left ankle Yes Intelligent filtering Show children concepts? No Empty input string or all combinations processed left foot No Process next word combination Yes Add matched term to list Match found? No Yes Replace matched terms from input string All combinations for current iteration finished? No

  16. Input string MRI ANKLE/FOOT WWO LT Empty or null string Return Remove stop words and special characters; convert to lower case mri ankle foot lt Replace known global abbreviations (e.g., LT with left) mri ankle foot left Determine modality and replace ankle foot left i) ankle foot left ii) ankle foot foot left ankle left Determine word combinations for input string in descending order Yes left foot left ankle Yes Intelligent filtering Show children concepts? No Empty input string or all combinations processed left foot No Process next word combination Yes Add matched term to list Match found? No Yes Replace matched terms from input string All combinations for current iteration finished? No

  17. Input string MRI ANKLE/FOOT WWO LT Empty or null string Return Remove stop words and special characters; convert to lower case mri ankle foot lt Replace known global abbreviations (e.g., LT with left) mri ankle foot left Determine modality and replace ankle foot left i) ankle foot left ii) ankle foot foot left ankle left Determine word combinations for input string in descending order Yes left foot left ankle Yes Intelligent filtering Show children concepts? No Empty input string or all combinations processed left foot No Process next word combination Yes Add matched term to list Match found? No Yes Replace matched terms from input string All combinations for current iteration finished? No

  18. Input string MRI ANKLE/FOOT WWO LT Empty or null string Return Remove stop words and special characters; convert to lower case mri ankle foot lt Replace known global abbreviations (e.g., LT with left) mri ankle foot left Determine modality and replace ankle foot left i) ankle foot left ii) ankle foot foot left ankle left Determine word combinations for input string in descending order Yes left foot left ankle Yes Intelligent filtering Show children concepts? No Empty input string or all combinations processed left foot No Process next word combination Yes Add matched term to list Match found? No Yes Replace matched terms from input string All combinations for current iteration finished? No

  19. Input string MRI ANKLE/FOOT WWO LT Empty or null string Return Remove stop words and special characters; convert to lower case mri ankle foot lt Regex based anatomy extraction: (?=.*\bank).*(?=.*\bfoo).*(?=.*\blef).* Replace known global abbreviations (e.g., LT with left) mri ankle foot left Determine modality and replace ankle foot left i) ankle foot left ii) ankle foot foot left ankle left Determine word combinations for input string in descending order Yes left foot left ankle Yes Intelligent filtering Show children concepts? No Empty input string or all combinations processed left foot No Process next word combination Yes Add matched term to list Match found? No Yes Replace matched terms from input string All combinations for current iteration finished? No

  20. Input string MRI ANKLE/FOOT WWO LT Empty or null string Return Remove stop words and special characters; convert to lower case mri ankle foot lt Replace known global abbreviations (e.g., LT with left) mri ankle foot left Determine modality and replace ankle foot left i) ankle foot left ii) ankle foot foot left ankle left Determine word combinations for input string in descending order Yes left foot left ankle Yes Intelligent filtering Show children concepts? No Empty input string or all combinations processed left foot No Process next word combination Yes Add matched term to list Match found? No Yes Replace matched terms from input string All combinations for current iteration finished? No

  21. Input string MRI ANKLE/FOOT WWO LT Empty or null string Return Remove stop words and special characters; convert to lower case mri ankle foot lt Replace known global abbreviations (e.g., LT with left) mri ankle foot left Determine modality and replace ankle foot left i) ankle foot left ii) ankle foot foot left ankle left Determine word combinations for input string in descending order Yes left foot left ankle Yes Intelligent filtering Show children concepts? No Empty input string or all combinations processed left foot No Process next word combination Yes Add matched term to list Match found? No Yes Replace matched terms from input string All combinations for current iteration finished? No

  22. Input string MRI ANKLE/FOOT WWO LT Empty or null string Return Remove stop words and special characters; convert to lower case mri ankle foot lt Replace known global abbreviations (e.g., LT with left) mri ankle foot left Determine modality and replace ankle foot left i) ankle foot left ii) ankle foot foot left ankle left Determine word combinations for input string in descending order Yes left foot left ankle Yes Intelligent filtering Show children concepts? No Empty input string or all combinations processed left foot No Process next word combination Yes Add matched term to list Match found? No Yes Replace matched terms from input string All combinations for current iteration finished? No

  23. Input string MRI ANKLE/FOOT WWO LT Empty or null string Return Remove stop words and special characters; convert to lower case mri ankle foot lt Replace known global abbreviations (e.g., LT with left) mri ankle foot left Determine modality and replace ankle foot left i) ankle foot left ii) ankle foot foot left ankle left Determine word combinations for input string in descending order Yes left foot left ankle Yes Intelligent filtering Show children concepts? No Empty input string or all combinations processed left foot No Process next word combination Yes Add matched term to list Match found? No Yes Replace matched terms from input string All combinations for current iteration finished? No

  24. Input string MRI ANKLE/FOOT WWO LT Empty or null string Return Remove stop words and special characters; convert to lower case mri ankle foot lt Replace known global abbreviations (e.g., LT with left) mri ankle foot left Determine modality and replace ankle foot left i) ankle foot left ii) ankle foot foot left ankle left Determine word combinations for input string in descending order Yes left foot left ankle Yes Intelligent filtering Show children concepts? No Empty input string or all combinations processed left foot No Process next word combination Yes Add matched term to list Match found? No Yes Replace matched terms from input string All combinations for current iteration finished? No

  25. Algorithm validation • Extracted 1604 production study descriptions from an academic institution • Used 1200 (~80%) for algorithm development • 404 used for testing • Accuracy of system was 99.94%. • (XR SACRUM COCCYX 2 VIEWS MIN was the false-negative)

  26. Results

  27. Limitations • Dataset is from one institution and therefore abbreviations and algorithms may need to be generalizable across institutions. • The reference taxonomy is not complete and represents only concepts encountered in the training set, as well as those included based on authors’ domain experience

  28. Key messages • Knowing the most specific body part of an imaging study is important for relevant prior study matching. • A regular expression based technique can be used to extract specific anatomy information from DICOM Study Description.

  29. Questions? Thusitha Mabotuwana thusitha.mabotuwana@philips.com

More Related