Studies on the Contours of Heart and Thorax Silhouettes

Abstract

In computer analysis of chest X-ray pictures, a large memory size and long processing time are usually necessary because of the sequential processing of the large amount of information contained in a picture. An approach which allows drastic data reduction is therefore pursued in order to develop practical systems for wide use. The contours of heart and thorax silhouettes must be intrinsically simple from the viewpoint of anatomy. However, the automated determination of the contours is not necessarily simple, because chest X-ray pictures have multi-gray levels and irrelevant silhouettes, or noises, overlap onto the contours.
Giving analytical bases to the developmental works on the automatic determination of the cardiothoracic ratios by the present author and his collaborators, discussed in this paper are :
1) a simplified cross-sectional model of the chest by introducing the concept of virtual thorax, followed by the definition of four characteristic points which are located on a density curve derived from a line scanning at right angle to the median line.
The basic concept, which is shown in analogy to the communication model, is that these points carry the information on the contours even if they might be disturbed by the noises.
2) description of the contours by Fourier series, which leads to the definition of the normalized complication measure R for the contours.
The measure R is defined as the ratio of the effective picture height to the interval of sampling points by which the contours can be approximated within specified errors. The results of statistical studies on R's with 19 random samples of chest X-ray pictures are presented.
3) the theoretical background of the algorithms for correcting errors in a data series of the characteristic points.
Two parameters, κ and η, are introduced. Also presented are the results of statistical studies on them. The criteria for error detection and correction are essentially derived from these parameters.