Impact of attenuation correction on the accuracy of FDG-PET in patients with abdominal tumors: a free-response ROC analysis

The aim of this study was to evaluate image quality and lesion detectabilit y with and without attenuation correction in patients with abdominal tumors , using a free-response receiver operating characteristic (FROC) methodolog y. Thirty-four patients with various abdominal tumors were evaluated (11 me n, 23 women, median age 48 years). Whole-body emission scans were performed 68 min (35-102 min) after intravenous injection of 4.3 MBq/kg fluorine-18 fluorodeoxyglucose (FDG). Images were reconstructed using the OS-EM algorit hm and corrected for attenuation either using postinjection singles transmi ssion (n = 27) or by calculation and body outline (n = 7). Total scan durat ion did not exceed 70 min. Studies were read independently by four observer s unaware of any clinical data, The uncorrected (UC) images were systematic ally read before the attenuation-corrected (AC) images. All studies were gi ven an image quality score ranging from 1 (unreadable) to 5 (excellent). Ea ch focus of increased activity was then localized and given a probability o f malignancy using a five-point scale. The average image quality score was similar for both UC and AC images. At the time of the positron emission tom ography (PET) scans, 127 lesions (63 liver metastases, 9 retroperitoneal le sions, 50 peritoneal or bowel lesions, and 5 pancreatic carcinomas) were re vealed by pathological or correlative studies. The areas under the FROG cur ves were consistently greater for AC images (range 0.8663-0.8867) than or U C images (range 0.7774-0.8613). Overall, the difference between the AC imag es and the UC images was significant (P = 0.019). In particular, correction for attenuation increased the sensitivity regardless of the location of th e lesions. In conclusion, correction for attenuation significantly improves the diagnostic accuracy of FDG-PET for abdominal staging of neoplasms, wit hout impairing the image quality.