Author: Ehsan Abadi, Njood Alsaihati, Steven T. Bache, Mridul Bhattarai, Cindy Marie McCabe, Francesco Ria, Ehsan Samei 👨🔬
Affiliation: Duke University, Center for Virtual Imaging Trials, Duke University, Duke University Health System, Clinical Imaging Physics Group, Department of Radiology, Duke University Health System 🌍
Purpose: To compare and contrast alternative methods including reader (in hominum), phantom (in phantasma), in vivo, and in silico methods deployed to assess the performance of photon counting (PCCT) compared to energy integrating CT (EICT).
Methods: Under IRB approval, 56 patients underwent chest CT on a PCCT and four EICT scanners from two different vendors on the same day. A physical phantom (Mercury 3.0), representing different adult population sizes, was imaged likewise. Five anthropomorphic phantoms (XCAT) were further virtually imaged at the same conditions using a validated, scanner-specific CT simulator (DukeSim). Four blinded radiologists evaluated image noise and resolution in terms of subsegmental bronchial wall definition (worst-best scale: 0-100). Image resolution and noise were measured in terms of Modulation Transfer Function’s 10% frequency (MTF-f10) and Global Noise Index (GNI) across all acquired images. Percentage differences between PCCT and EICT images were calculated for observer ratings and image quality across all images.
Results: On average, radiologists rated the patient PCCT images 13.6% ± 20.5% sharper and 21.4% ± 29.0% less noisy compared to the EICT ones. MTF-f10 was sharper in the PCCT images by 24.6% ± 20.8%, 5.6% ± 13.2%, and 255.4% ± 42.0% in the in vivo, in silico, and in phantasma data, respectively. GNI was lower in the PCCT images by 212.4% ± 165.8%, 54.6% ± 13.5%, and 143.0% ± 50.0% in the in vivo, in silico, and in phantasma data, respectively.
Conclusion: Observer studies represent the most relevant approach when evaluating the image quality performance of CT technology. However, they are time-consuming and routinely unfeasible. Alternative methods offer different levels of realism, practicality, and value. Phantom measures are most practical while in vivo measures are best for clinical surveillance and tracking purposes. In silico data was found to offer the closest representation of image quality to observer ratings.