Premium
Development of a computerized method for identifying the posteroanterior and lateral views of chest radiographs by use of a template matching technique
Author(s) -
Arimura Hidetaka,
Katsuragawa Shigehiko,
Li Qiang,
Ishida Takayuki,
Doi Kunio
Publication year - 2002
Publication title -
medical physics
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 1.473
H-Index - 180
eISSN - 2473-4209
pISSN - 0094-2405
DOI - 10.1118/1.1487426
Subject(s) - radiography , template matching , template , matching (statistics) , similarity (geometry) , digital radiography , medicine , correlation , computer science , radiology , artificial intelligence , nuclear medicine , medical physics , image (mathematics) , mathematics , pathology , geometry , programming language
In picture archiving and communications systems (PACS) or digital archiving systems, the information on the posteroanterior (PA) and lateral views for chest radiographs is often not recorded or is recorded incorrectly. However, it is necessary to identify the PA or lateral view correctly and automatically for quantitative analysis of chest images for computer‐aided diagnosis. Our purpose in this study was to develop a computerized method for correctly identifying either PA or lateral views of chest radiographs. Our approach is to examine the similarity of a chest image with templates that represent the average chest images of the PA or lateral view for various types of patients. By use of a template matching technique with nine template images for patients of different size in two steps, correlation values were obtained for determining whether a chest image is either a PA or a lateral view. The templates for PA and lateral views were prepared from 447 PA and 200 lateral chest images. For a validation test, this scheme was applied to 1,000 test images consisting of 500 PA and 500 lateral chest radiographs, which are different from training cases. In the first step, 924 (92.4%) of the cases were correctly identified by comparison of the correlation values obtained with the three templates for medium‐size patients. In the second step, the correlation values with the six templates for small and large patients were compared, and all of the remaining unidentifiable cases were identified correctly.