Phase congruency parameter estimation and discrimination ability in detecting lung disease chest radiograph

Abstract

The conventional chest radiograph remains a widely tool in the diagnosis of lung diseases even to the present day. Current methods or algorithms for disease detection focus on the discrimination between normal images and images with signs of disease involving chest radiograph. This paper proposed a novel algorithm to solve the difficult problem of discriminating two similar diseases, pulmonary tuberculosis (PTB) and lobar pneumonia (PNEU) using phase congruency. The phase congruency PC(x) parameter estimation was studied to obtain the best PC(x)-values that has the ability to differentiate between normals, PTB and PNEU. Eight texture measures of PC(x) values were then investigated as global measures for differentiation of diseases. All eight of these texture measures were found to have univariate normal distributions which allowed the statistical discriminant function, D(x), to select the best texture measures. The homogeneity texture measure gave the best discrimination for PTB and PNEU with Type 1 Error of 0.1 while the Type II Error of 0.15.