Meeting Banner
Abstract #3013

An Image Searching Engine to Utilize Past Clinical Data for the Future Diagnosis

Andreia V. Faria1, Shoko Yoshida1, Kenichi Oishi1, Kanako Sato1, Argye Hillis2, MIchael I. Miller3, Susumu Mori1

1Radiology, Johns Hopkins University, Baltimore, MD, United States; 2Neurology, Johns Hopkins Hospital, Baltimore, MD, United States; 3Biomedical Engineering, Johns Hopkins University, Baltimore, MD, United States

We report our attempt to develop a technology to structurize image features and facilitate image searching. The structurization was based on automated parcellation of the entire brain into 211 structures using T1-WIs and high-dimensional normalization method. We tested if the structurized anatomical data actually captured the anatomical features in a population with atrophy at different degrees and locations by comparing the results with trained clinicians evaluation. We explored the data and tested individual classifications using PCA and discriminant analysis. The structurization of image data through image-vector conversion was effective, and provides opportunities to mine clinical database for medical decision support

Abstract PDF Presentation Video

Keywords

accuracy actually agreement allow although among analyses anatomic anatomical anatomy aphasia appearance applications asked assignment atlas atrophy audience automated biomedical brain calls capture characterized clinical clinicians closer common component controls conversion crucial currently database decision deep deformation described develop diagnosis diagnostic dimensional direct distant distinguish document early educational engine entire evidences examine existing extension facilitate features findings frontal functional future grants gray gross hand hospital human ideal identification impairment increasing indicates individual initial integrated integration integrative inter johns judgment loading lobe major make making mapping medical medicine metric mild miller model natural neighboring occipital opportunities parietal particularly past patient pattern perceived perfect phenotype phenotypes plot population power primary principal progressive quantitative radiological radiologists radiology raters reach revealed score scores search searched searching seems segregation semantic share stages structured structures subjective superficial support symptoms systematic target technology template temporal text third tools trained understandable various vary vector vectors virtually visual visually white