A preliminary study on the knowledge-based delineation of anatomical structures for whole body PET-CT studies

Abstract

PET-CT imaging has shown its superiority in the clinical management of cancer. The markedly increased amount of imaging data have given rise to the development of computer-aided diagnosis (CAD) to aid the clinician in the interpretation of large volumes of data. The delineation of anatomical structures is one of the major components of CAD. Currently, the majority of segmentation methods are focused on the segmentation of organs and tissues using high-contrast anatomical images such as high-dose CT with injected contrast agent. However, typically low-dose CT protocol without the use of contrast agent are used in PET-CT studies, which leads to low-contrast CT images with a relatively high level of noise. This study investigated the potential of using information extracted from the co-registered PET-CT data in the segmentation of anatomical structures. A preliminary knowledge-based system was developed to process eight clinical PET-CT studies for lung cancer. The results of qualitative and quantitative analysis demonstrate the efficiency of incorporating the information derived from co-registered structural and functional images in the segmentation of anatomical structures for whole body PET-CT studies. It also implies that the methods relying on the HU value, like thresholding, are incapable of accurately delineating those organs suffering from high-level noise with unclear boundary. Further investigation using advanced technologies are warranted to achieve accurate segmentation for PET-CT imaging.