Deformations, patches, and discriminative models for automatic annotation of medical radiographs

  • Authors:
  • Thomas Deselaers;Hermann Ney

  • Affiliations:
  • RWTH Aachen University, Computer Science Department, Human Language Technology and Pattern Recognition Group, Aachen, Germany;RWTH Aachen University, Computer Science Department, Human Language Technology and Pattern Recognition Group, Aachen, Germany

  • Venue:
  • Pattern Recognition Letters
  • Year:
  • 2008

Quantified Score

Hi-index 0.10

Visualization

Abstract

In this paper, we describe three different methods for the classification and annotation of medical radiographs. The methods were applied in the medical image annotation tasks of ImageCLEF in 2005, 2006, and 2007. Image annotation can be used to access and find images in a database using textual queries when no textual image description is available. One of the methods is a non-linear model taking into account local image deformations to compare images which are then classified using the nearest neighbour decision rule. The other two methods use local image descriptors for a bag-of-features approach. The bags of local image features are classified using discriminative classifiers. Our methods performed best in the 2005 and 2006 evaluations and second best in 2007.