Can photon-counting CT improve PET/CT’s quantitative accuracy?

Introduction: The latest generation of CT scanners with photon-counting detectors (PCT-CT), a new advance relative to energy-integrating detectors (EID-CT), are not incorporated into commercially available PET/CT systems. It is unknown whether spectral information from a single low-dose acquisition may lead to more accurate estimation of PET linear attenuation coefficients (LAC). This work, based on phantom scans performed on clinical systems, asks if the accuracy in LAC can lead to improved quantitative accuracy in PET/CT images.

Methods: Ten one-liter plastic bottles were filled with water and various concentrations of salt (simulating blood, 40 g/l), iodine contrast medium (23 and 11.5 mgI/l), K₂HPO₄ (bony tissue, 500 and 125 g/l), and polystyrene beads (lung). ¹⁸F-FDG was added to the bottles measured using a dose calibrator (2.25 to 2.52 kBq/ml, and one with no activity). The bottles were arranged in a tray, comprising a 41x16x16 cm³ phantom. A Siemens Biograph mCT 128 scanner (with EID-CT) performed PET/CT scans with the phantom in two orientations, 41 and 16 cm wide, simulating whole-body scanning and head scanning. The phantom was scanned again in the same configurations on a NAEOTOM Alpha scanner (Siemens Healthineers). PET images were reconstructed two ways: first, with LAC based on normal EID-CT (140 kV tube voltage) from the mCT scanner; and, second, with LAC in which PCT-CT data were first converted to virtual monoenergetic images (VMI) and then spatially co-registered with the EID-CT images. EID-CT image values in Hounsfield units were converted to LAC with the standard bilinear model. VMI image values were assumed to be proportional to electron density (ED) and were converted to LAC with a simple linear model. Apart from the calculation of LAC, the same reconstruction method was used for all PET reconstructions. In addition to the PET/CT experiments, a Sun Nuclear model 1472 electron density phantom was scanned with both EID-CT and PCT-CT. This phantom included 16 inserts of known ED. In their chemical composition, the inserts simulated body tissues including bone, blood, and blood with iodine contrast. LAC estimated from EID-CT and PCD-CT were compared to ground truth values from the manufacturer. The EID-CT comparison included LAC arising from scans at 80 kV and 140 kV.

Results: In the PET phantom when normal EID-CT based LAC were used, reconstructed Bq/ml values were within 5% of assay values in 10 of 16 measurements. When PCT-CT based LAC were used, 14 of the 16 measurements were within 5%, indicating improved accuracy with PCT-CT, though the accuracy of the assay itself was not known. That analysis included the phantom’s body and head-like orientations and excluded a bottle with no activity and a bottle with polystyrene beads. The average absolute errors, when reconstructed PET images were compared to the assay, were 5.3% (EID-CT) and 2.8% (PCT-CT). Absolute differences between the body-like and head-like orientations were on average 3.9% (EID-CT) and 3.2% (PCT-CT), a comparison that did not rely on the assumption of an accurate assay. In CT scans of the ED phantom, LAC of inserts that contained iodine deviated by up to 20% on EID-CT images, but were within 2% of known values when calculated from VMI from PCT-CT. It should be noted that both parts of this study used concentrations of iodine and bone surrogate that were high relative to clinical PET/CT. It should also be noted that VMI are currently available in state-of-the-art EID-CT, but this requires dual-energy measurements with split filters (e.g. Siemens TwinBeam) or with two spiral acquisitions at different kV settings

Conclusions: Attenuation correction in future PET/CT scanners may be quantitatively more accurate if PCT-CT is used instead of EID-CT. In this phantom study, PET/CT errors were on average reduced from 5.3% to 2.8%.

• Download figure
• Open in new tab
• Download powerpoint