Liver Lesion Volumetry Using Fully Automated Lesion Segmentation- Photon-Counting CT Versus Dual Energy CT
Abstract
Purpose
To study the accuracy and reproducibility of liver lesion volume measurements using fully automated lesion segmentation at virtual monochromatic imaging (VMI) with photon-counting and dual energy CT (PCCT and DECT).
Methods
A custom-made anthropomorphic liver phantom (size: 25 x 32.5 cm2) was scanned using GE Revolution DECT (GSI, ASIR-V=40%) and Siemens Naeotom PCCT (quantum+, QIR=2), each at 120 kV with three repeats at CTDIvol = 14 mGy. The phantom liver parenchyma was perfused with iodine of 0, 0.68, and 2.38 mg/mL at different slabs to simulate a multi-phase exam. The built-in lesions include iodine and low-density liver, iron oxide, fat, and cyst. 28 visually discernable ellipsoidal or lobular lesions, with nominal contrast from 10 HU to 160 HU and volumes from 0.21 – 8.57 mL, were automatically segmented at 50 – 100 keV using Siemens SyngoVia MM Oncology. The measured volumes were compared to the known values. To account for both accuracy and reproducibility at each keV, F-beta score (F10), with beta equal to 10, was used to place more weight to reproducibility.
Results
F10 scores for the iron oxide, cyst, and fatty lesions ranged from 0.94- 0.99 across 50 -100 keV except at 50 keV where the score of the cyst was 0.84 with DECT. F10 scores for iodinated and low liver density lesions were from 0.70- 0.92 with the maxima of 0.85 and 0.92 at 60 keV, for DECT and PCCT, respectively. In comparison to the ground-truth volumes, the root-mean-square (R.M.S.) errors for all lesions at 60 keV were 12.9% and 6.3%, for DECT and PCCT, respectively.
Conclusion
PCCT with fully automated lesion segmentation resulted in substantially better volume accuracy and reproducibility. 60 keV was found to be optimal for both PCCT and DECT. The R.M.S. error of 6.3% from PCCT suggests reliable lesion volume measurements can be made.