A dedicated breast-PET/CT scanner: Numerical observer study of lesion detection

Abstract

Background: Dedicated, breast-specific positron emission tomography (BPET)-cone-beam computed tomography (BPET/CT) systems have been developed to improve detection and diagnosis of cancer in women with indeterminate mammograms caused by radiodense breasts. The absorption of X-rays that often vexes mammography in this subset of women does not affect the detection of the high energy annihilation photons used in PET. PET imaging of the breast, however, is subject to limitations caused by their comparatively low spatial resolution (∼2 mm) and often moderate radiotracer uptake in lesions.

Purpose: The purpose of this investigation is to explore the PET-based lesion detection capabilities of a BPET/CT scanner developed by the Department of Radiology Instrumentation group at West Virginia University.

Methods: The PET component of the system consists of a rotating pair of 96 × 72 arrays of 2 × 2 × 15 mm³ LYSO scintillator elements. The cone-beam-CT component utilized a pulsed X-ray source and flat panel detector operated in portrait orientation. The density maps created by the CT scanner were used to correct the BPET data for photon attenuation and Compton scattering. The nonuniform uptake of ¹⁸ F-fluorodeoxyglucose (FDG) in normal breast tissue was emulated in a specially designed phantom consisting of an acrylic cylinder filled with a mixture of acrylic beads and liquid containing FDG. FDG-avid lesions were simulated with agar spheres (3, 4, 6, 8, and 10 mm diameters) containing vary amounts of FDG to produce target-to-background ratios (TBR) of 6:1, 8:1, and 10:1. The spheres also contained X-ray contrast agent to make even the smallest ones readily visible in CT images. Positions of all the lesions were identified in the CT images. These positions were used to extract signal present and signal absent sub-images from the PET images. The sub-images were then input to software that calculated areas-under-the-curve for two numerical model observers (Laguerre-Gauss channelized Hotelling observer and non-prewhitening matched filter).

Results: The results showed that the smallest detectable lesion with this system is no smaller than ∼3 mm in diameter with a TBR of 6:1. Simulated lesions with diameters of 4 mm and greater were calculated to have good to excellent likelihood of detection for all TBRs tested.

Conclusion: The results from this investigation identified the detectability capabilities and limitations for a dedicated breast-PET/CT scanner. Its ability to detect relatively small simulated FDG-avid breast lesions for a range of TBRs indicates its potential for clinical application. Finally, the study used methodologies that could be applied to a detectability assessment of other PET/CT scanners.

Keywords: CT; PET; breast; numerical observer.