The purpose of this work is to create a rigorous method of segmenting PET images using an automated iterative technique. To this end a phantom study employing spherical targets was used to determine local (slice specific) threshold levels which produce correct cross-sections based on the contrast between target and background. Numerous target to background activity concentration ratios were investigated but found to have minimal effect in comparison to the influence of target size. Functions were fit to this data and used to construct an iterative threshold segmentation algorithm. In all cases this approach yielded convergence within ten iterations. Iterative threshold segmentation was applied using both an axial and tri-axial approach to the spherical targets and also to two irregularly shaped volumes. Of these two approaches, the tri-axial method proved less susceptible to image noise and better at dealing with partial volume effects at the interface between target and background. For comparative purposes, single thresholds of 28% and 40% were also applied to the spherical data sets. The tri-axial iterative method was found capable of delineating cross sections with areas greater than 250 mm2 to within the maximum resolution possible (1 pixel width). Cross sections of less than 250 mm2 in area were resolved by the tri-axial method to within 2 pixel widths of their true physical extent. Local contrast based iterative threshold segmentation shows promise as a method of rigorously delineating PET target volumes with good accuracy subject to the limitations imposed by the image resolution which currently characterizes this modality.
Current radiation therapy techniques, such as intensity modulated radiation therapy and three-dimensional conformal radiotherapy rely on the precise delivery of high doses of radiation to well-defined volumes. CT, the imaging modality that is most commonly used to determine treatment volumes cannot, however, easily distinguish between cancerous and normal tissue. The ability of positron emission tomography (PET) to more readily differentiate between malignant and healthy tissues has generated great interest in using PET images to delineate target volumes for radiation treatment planning. At present the accurate geometric delineation of tumor volumes is a subject open to considerable interpretation. The possibility of using a local contrast based approach to threshold segmentation to accurately delineate PET target cross sections is investigated using well-defined cylindrical and spherical volumes. Contrast levels which yield correct volumetric quantification are found to be a function of the activity concentration ratio between target and background, target size, and slice location. Possibilities for clinical implementation are explored along with the limits posed by this form of segmentation.
Incorporation of positron emission tomography (PET) data into radiotherapy planning is currently under investigation for numerous sites including lung, brain, head and neck, breast, and prostate. Accurate tumor‐volume quantification is essential to the proper utilization of the unique information provided by PET. Unfortunately, target delineation within PET currently remains a largely unaddressed problem. We therefore examined the ability of three segmentation methods—thresholding, Sobel edge detection, and the watershed approach—to yield accurate delineation of PET target cross‐sections. A phantom study employing well‐defined cylindrical and spherical volumes and activity distributions provided an opportunity to assess the relative efficacy with which the three approaches could yield accurate target delineation in PET. Results revealed that threshold segmentation can accurately delineate target cross‐sections, but that the Sobel and watershed techniques both consistently fail to correctly identify the size of experimental volumes. The usefulness of threshold‐based segmentation is limited, however, by the dependence of the correct threshold (that which returns the correct area at each image slice) on target size.PACS numbers: 87.58.Fg, 87.57.Nk
scite is a Brooklyn-based organization that helps researchers better discover and understand research articles through Smart Citations–citations that display the context of the citation and describe whether the article provides supporting or contrasting evidence. scite is used by students and researchers from around the world and is funded in part by the National Science Foundation and the National Institute on Drug Abuse of the National Institutes of Health.