A physician-in-the-loop content-based image retrieval system for medical image databases
Abstract
It is now recognized in many domains that content-based image retrieval (CBIR) from an image database cannot be carried out by using completely automated approaches. One such domain is medical radiology for which the clinically useful information in an image typically consists of gray level variations in highly localized regions of the image. Currently, it is not possible to extract these regions by automatic image segmentation techniques. To address this problem, we have implemented a “physician-in-the-loop” approach in which a physician delineates the pathology bearing regions (PBRs) and a set of anatomical landmarks in the image when the image is entered into the database. A suite of computer vision algorithms are applied to the region thus marked to extract the attributes to determine the presence or the absence of the various perceptual categories which the physicians claim to use for classifying different disease categories. Subsequently we analyze the discriminatory power of the perceptual categories by applying a statistical tool called MANOVA. The image attributes that are found to be maximally discriminatory for distinguishing among the different pathology classes are grouped with the PBRs' spatial attributes into a special data structure called a lobular feature set. Applying a multi-dimensional hashing approach, the system then assigns an overall index to each image based on its lobular feature sets. This index is used for archiving the image and its retrieval. Our system also features a graphical user interface for query formulation and retrieval results visualization.
Degree
Ph.D.
Advisors
Kak, Purdue University.
Subject Area
Computer science|Electrical engineering|Biomedical research
Off-Campus Purdue Users:
To access this dissertation, please log in to our
proxy server.