imaging informatics VI

  1. CAD
    • computer-assisted diagnosis system
    • what/where is the disease?
    • automated diagnosis using qualitative features
    • automated diagnosis using quantitative features
    • automated localization and annotation of lesions
  2. CAD: automated diagnosis using qualitative features example
    predict diagnosis of the patient based on the imaging abnormalities using Bayesian reasoning
  3. CAD: automated diagnosis using quantitative features example
    predicing the type of lung cancer (recurring or non-recurring) using quantitative image texture features
  4. CAD: automated localization and annotation of lesions example
    • image labeling to deduce structures affected by the disease
    • atlas image (anatomy overlap) and fMRI image (functional map)
  5. imaging applications: patient treatment planning
    • use images to plan approach to treatment
    • surgical planning: use images to predict best surgical path
    • given 3D volumetric imaging data
    • given knowledge of connectivity
    • link images to knowledge
    • predict best path to minimize tissue/organ damage
    • requires anatomic reasoning
  6. imaging applications: treatment response assessment
    • analyze images to determine if patients are responding to treatment
    • assessed on longitudinal image (repeated imaging studies obtained over time)
    • approach: detect change in disease (change in size or other characteristic)
    • reasoning task: analyze lesion measurements over time to determine if treatment is working
    • needs a standard criteria for describing assessment - RECIST
  7. RECIST
    • response evaluation of criteria in solid tumors
    • scores for tumor progression
  8. Challenges with computerized assisted detection/diagnosis
    • models of disease: need good model relating image features to diagnosis, need good image features as inputs to models
    • data access: privacy issues, etc.
    • explanation: users don't like black boxes
    • integration: with clinical workflow
    • lack of tools/standards: for accessing images to build these systems
  9. Methods of CAD
    • rule based
    • statistical/machine learning models
  10. CAD: rule-based methods
    use OWL to query ontologies and reason with images to make rule-based DSS
  11. CAD statistical/machine learning models
    • quantitative features: many machine learning models, logistic regression, lasso, SVM
    • qualitative (semantic) features: Bayesian networks, image interpretation involves probability
  12. Bayesian networks: causal and inferential reasoning
    • causal: given the disease, what is the probability of observing the findings?
    • inferential: the probability of disease given the findings
  13. OWL
    • has explicit semantics - machines can understand
    • enables automated computer reasoning
    • part of Semantic Web technology stack
    • lets us define ontology classes
    • we specify necessary and sufficient conditions for class membership
    • specified using formal logic expressions
    • things meeting these conditions will be classified to the class
    • generic software program called an automatic classifier does the work
    • lets us describe classes in ontologies, rather than just naming them
  14. content based image retrieval: querying knowledge representations
    • atlases: to transfer knowledge to your image, what anatomical structure is at point X?, use image registration to overlap with atlas
    • ontologies: find classes - is "chest" in RadLex?, find attribute values for a given class - what are synonyms for "chest"?, traverse relations - query expansion, answer questions
  15. content based image retrieval: querying images
    • search for images: find images showing a mass in the liver
    • summarize information in images: how much is the cancer size changing across multiple CT scans?
    • make decisions based on images: is the patient's cancer responding well to treatment?
  16. How can we query images?
    • Querying DICOM with text mining
    • Querying AIM with API
    • Querying RDF annotations with SPARQL
  17. Querying DICOM with text matching: what?
    • string matching in DICOM headers
    • iterate through DICOM files, grab field values, and do string matching
    • several libraries available
  18. Querying DICOM with text matching: types of questions we can answer
    • get a list of patients in my database
    • get the studies that are of particular modality (all CT scans)
    • get images of Mary Jane
  19. Querying DICOM with text matching: CAN'T answer these questions
    • find all the CT images of the chest
    • find similar images
    • how much is the cancer size changing across multiple CT exams
  20. API
    • application programming interface
    • gives (controlled) access to database or web resource
  21. ePAD
    • electronic physician annotation device
    • web-based image viewer and annotator
    • AIM-compliant AIM annotation of images
    • API for querying semantic image features
    • can query AIM files using ePAD's REST API
    • generates AIM files that contain semantic features
  22. Allen Brain Atlas API
    image annotations
  23. Querying AIM with API: questions we can answer
    • find image showing a mass in the liver: query AIM annotations for ImagingObservation="mas" on images
    • find similar images: query semantic annotations on images to find the images that share semantic annotations
    • how much is the cancer size changing across multiple CT exams: query AIM annotations across multiple image acquired over time
  24. Querying AIM with API: question we CAN'T answer
    is the patient's cancer responding well to treatment
  25. SPARQL
    simple protocol and RDF query language
  26. Query RDF annotations with SPARQL: question that can be answered
    • those from other methods
    • is the patient's cancer responding well to treatment: computerized reasoning with image annotations
    • find CT images of the chest
  27. image querying method to answer: get a list of patients in my database
    querying DICOM with text matching
  28. image querying method to answer: get the studies that are of a particular modality (all CT scans)
    querying DICOM with text matching
  29. image querying method to answer: get images of Mary Jane
    querying DICOM with text matching
  30. image querying method to answer: find images showing a mass in the liver
    Querying AIM with API
  31. image querying method to answer: find similar images
    Querying AIM with API
  32. image querying method to answer: how much is the cancer size changing across multiple CT exams
    Querying AIM with API
  33. image querying method to answer: is the patient's cancer responding well to treatment?
    Query RDF annotations with SPARQL
Author
tulipyoursweety
ID
350192
Card Set
imaging informatics VI
Description
imaging informatics
Updated