Characterisation of radiotherapy planning volumes using textural analysis

Computer-based artificial intelligence methods for classification and delineation of the gross tumour volume (GTV) on computerised tomography (CT) and magnetic resonance (MR) images do not, at present, provide the accuracy required for radiotherapy applications. This paper describes an image analysis method for classification of distinct regions within the GTV, and other clinically relevant regions, on CT images acquired on eight bladder cancer patients at the radiotherapy planning stage and thereafter at regular intervals during treatment. Statistical and fractal textural features (N=27) were calculated on the bladder, rectum and a control region identified on axial, coronal and sagittal CT images. Unsupervised classification results demonstrate that with a reduced feature set (N=3) the approach offers significant classification accuracy on axial, coronal and sagittal CT image planes and has the potential to be developed further for radiotherapy applications, particularly towards an automatic outlining approach.