A method for the recognition of inflamatory nodules from chest X-ray CT images

Abstract

In this paper, we propose a method to recognize inflammatory nodules from thin slice chest CT images. These nodules can be visually decided their benignancy using medical observations without biopsy or follow-up. These medical observations were quantified to be used in CAD (Computer Aided Diagnosis) by six features: nodules features such as ratio of linear part in the bounding, size, shape complexity, the standard deviation of CT values inside it, two features between nodules and chest walls (Distance and existence of interlobular septum). In order to recognize inflammatory nodules, we use the logistic regression model. The method was applied to eighty-three small and solid nodules including forty-four inflammatory nodules. The accuracy of the recognition was seventy-eight percent. Furthermore, we tested hypothesis for the significance of the features. It was confirmed that the two features (Size and standard deviation of CT values) are significant.