Interactive method to extract tumor regions from chest X-ray CT images and its application to benign/malignant discrimination

Abstract

This paper proposes a method that extracts interactively the lung tumor region from the chest X-ray image for detailed examination, while keeping the boundary shape information. Several methods have been proposed to extract the position and the rough shape of the tumor from the CT image; however, they were developed for screening purposes, and are not suited to the benign/malignant discrimination, where the tumor shadow itself should be used as the region for calculation of the features. In addition, there has been little development of a system that supports the diagnosis by extracting the lung tumor region in collaboration with the medical doctor. With this background, the authors developed an extraction method by an interactive operation, while preserving the boundary shape of the tumor. The method is mostly based on thresholding and the distance transformation. The method is applied to 78 cases of practical CT images, and the validity of the extracted tumor shape is verified. Then, the benign/malignant discrimination of the tumor is attempted by determination of various feature parameters from the extracted tumor region, and it is shown that the correct discrimination rate is improved. © 2005 Wiley Periodicals, Inc. Syst Comp Jpn, 36(8): 90–101, 2005; Published online in Wiley InterScience (www.interscience.wiley.com). DOI 10.1002/scj.20177