Texture analysis of 3D bladder cancer CT images for improving radiotherapy planning

Abstract

At present no single texture analysis approach can provide automatic classification to the accuracy required for radiotherapy applications. The method presented was developed to classify areas within the gross tumor volume (GTV), and other clinically relevant regions, on computerized tomography (CT) images. For eight bladder cancer patients, CT information was acquired at the radiotherapy planning stage and thereafter at regular intervals during treatment. Textural features (N=27) were calculated on regions extracted within the bladder, rectum and a region identified as clinically relevant. The sequential forward search (SFS) method was used to reduce the feature set (N=3). The results demonstrate the significant sensitivity of the reduced feature set for classification of any orthogonal CT image and the potential of the approach for radiotherapy applications.