2016 Winter Project Week/Projects/BatchImageAnalysis

From NAMIC Wiki
Revision as of 15:54, 8 January 2016 by Lauren (talk | contribs)
(diff) ← Older revision | Latest revision (diff) | Newer revision → (diff)
Jump to: navigation, search
Home < 2016 Winter Project Week < Projects < BatchImageAnalysis
3D SIFT Lung Features

Key Investigators

  • Kalli Retzepi (MGH)
  • Yangming Ou (MGH)
  • Matt Toews (ETS)
  • Lilla Zollei (MGH)
  • Lauren O'Donnell (BWH)
  • Steve Pieper (BWH)
  • Sandy Wells (BWH)
  • Randy Gollub (MGH)

Project Description

Objective Approach and Plan Progress and Next Steps
  • Run feature detection code over a collection of medical images pulled from PACS
  • Investigate a collection of ADC maps of neonates (diffusion MR)
  • Patients labeled with age and health status (normal, mildly abnormal, severely abnormal)
  • Use 3D SIFT code to see if health status can be detected in images
  • (if time) try text analysis of radiology reports
  • Use deidentified cohort of neonate images collected from MGH
  • Install data and software on AWS, try StarCluster
  • Explore visualization options
  • (if time) integrate image features with analysis of radiology report text

Algorithm

  • feature extraction (20 seconds per image)
  • Feature matching O(log N) indexing (< 1 second per image)
  • 3D SIFT-Rank code (Windows, Linux, Max) and read me

http://www.matthewtoews.com/fba/featExtract1.5.zip

Result Baseline HIE classification rate: 73%, leave-one-out moderate vs normal.


Data 231 subjects, Apparent Diffusion Coefficient (ADC) images.

Features Extracted in ADC MRI Volume

Image_%282%29.png