A.B. Barclay,
R.L Eisner, and E.V.R. DiBella.
Carlyle Fraser Heart Center/Crawford Long Hospital of Emory University and Georgia Institute of Technology, Atlanta, GA.
Abstract
The analytical thorax phantoms used in PET and SPECT algorithm evaluation may not accurately reflect the size and shape of a real patient population. We have constructed a new thorax attenuation database from 167 clinical PET rotating-rod transmission scans on patients ranging in weight from 85 to 245 pounds.
Each patient entry consists of demographic information, left-ventricle (LV) orientation information, a filtered back-projection (FBP) reconstruction of attenuation coefficients (AC), and a parameterization of the patient's body and lung boundaries. The body and lung boundaries were determined from the two-dimensional AC images using a thresholding technique. The boundaries were used to classify patients and to warp the AC images to template boundaries that could then be used to build AC image models.
An infinite variety of attenuation models can be built by classifying patients into groups based on parameters from the database. Classifications can be based on parameters such as weight, sex, LV orientation, lung volume, or lung position. Models have been built from six classifications of patients: the average of all 167 patients, all males, all females, 5 patients with prominent breasts, 5 patients with thin chest walls, and 5 patients with thick chest walls. The models not only supply information about the average boundaries and AC of the groups, but also the standard deviations.
Using boundary fitting and image warping techniques, realistic model shapes can be filled with real attenuation coefficients. The new thorax model database will be used to illustrate attenuation artifacts, perform Monte Carlo simulations and test algorithms for PET and SPECT attenuation compensation.
Andrew B. Barclay / abb@nuccard.eushc.org Last updated 6/20/95