A new texture analysis method for classification of interstitial lung abnormalities in chest radiography

This paper presents a new method called ´Score-Block operation´, which is originally from the general image retrieval concept, with k nearest neighbour classifier for automated categorization of interstitial pattern in chest radiographs. Before this method is applied on a chest radiograph, the lung field is extracted from the image and image feature vectors for each region of interest are obtained using power spectrum and Quasi-Gabor filter. There are three problems to solve: padding of the border, quantification of the interstitial abnormalities of chest radiograph and ribs and vessel shadows of lung field. The Score-Block operation solves the first and second problems. The third problem is solved by two partition image databases (PIDB), which are normal partition image database and abnormal partition image database. PIDBs consist of all normal and abnormal partition images including other superimposed organs, so our system is not sensitive to the ribs and vessels shadow.