Academic radiology · 2026
Small-pixel Acquisition Unmasks "Bucket Effects" of Iterative Reconstruction in Ultra-High-Resolution Photon-Counting CT.
СтатьяНазвание и аннотация: Перевод в очереди
Обработка статьи · этапы, ошибки и управление
Загрузка состояния…
Ошибки и ограничения
История · было → стало
Состояние обновляется каждые 5 секунд, пока блок открыт. Процент показывается только там, где известен объём работы. Пауза останавливает продолжение этапа; текущий запрос к источнику может завершиться.
Huflage Henner, Wech Tobias, Patzer Theresa Sophie, Bestler Caroline, Pannenbecker Pauline, Gruschwitz Philipp, Weber Anna, Kunz Andreas Steven, Bley Thorsten Alexander, Müller Lukas, Grunz Jan-Peter
Аннотация
RATIONALE AND OBJECTIVES: Ultra-high-resolution (UHR) examinations do not entail a dose burden in photon-counting CT (PCCT), since no post-patient filter is required to narrow the detector aperture. While this configuration is known to improve spatial sampling ("small pixel effect"), the interaction between reconstruction kernels and iterative reconstruction (IR) remains incompletely understood. This study investigated how smaller detector pixels influence signal-to-noise behavior and subjective image quality in midface PCCT.
MATERIALS AND METHODS: Fourteen cadaveric heads were examined using both UHR (120 × 0.2 mm) and standard collimation (144 × 0.4 mm) on twelve dose levels (0.08-10 mGy) with a first-generation cadmium-telluride PCCT scanner. UHR data were reconstructed with three dedicated head kernels with spatial frequencies of 12.8, 21.0, and 27.9 line pairs/cm at 10% of the modulation transfer function (Hr64, Hr76, and Hr84). Only Hr64 and Hr76 could also be used for standard-resolution data. Signal-to-noise ratios (SNR) were computed among all datasets. Subjective image quality was assessed by nine radiologists using a forced-choice pairwise comparison tool.
RESULTS: Comparing reconstructions with the sharp Hr76 kernel, SNR was higher in UHR than in standard-resolution datasets on each dose level (all p<0.001). In contrast, no significant difference was established between reconstructions with the intermediate Hr64 kernel (all p≥0.300). Notably, in UHR acquisitions, increasing kernel sharpness did not lead to the expected SNR penalty: Hr76 demonstrated comparable or numerically higher SNR than Hr64 across dose levels (all p≥0.050). This counterintuitive pattern suggests a non-uniform, "bucket-like" interaction between reconstruction kernels and IR. Subjective analysis showed substantial inter-reader agreement (Fleiss κ=0.711) and consistent preference for sharper UHR reconstructions, even when acquired at lower radiation dose.
CONCLUSION: In midface PCCT, smaller detector pixels enable a measurable SNR advantage when matched for reconstruction sharpness. In addition, kernel-dependent interactions with IR may lead to non-monotonic noise behavior, allowing sharper reconstructions without the expected noise penalty.
MATERIALS AND METHODS: Fourteen cadaveric heads were examined using both UHR (120 × 0.2 mm) and standard collimation (144 × 0.4 mm) on twelve dose levels (0.08-10 mGy) with a first-generation cadmium-telluride PCCT scanner. UHR data were reconstructed with three dedicated head kernels with spatial frequencies of 12.8, 21.0, and 27.9 line pairs/cm at 10% of the modulation transfer function (Hr64, Hr76, and Hr84). Only Hr64 and Hr76 could also be used for standard-resolution data. Signal-to-noise ratios (SNR) were computed among all datasets. Subjective image quality was assessed by nine radiologists using a forced-choice pairwise comparison tool.
RESULTS: Comparing reconstructions with the sharp Hr76 kernel, SNR was higher in UHR than in standard-resolution datasets on each dose level (all p<0.001). In contrast, no significant difference was established between reconstructions with the intermediate Hr64 kernel (all p≥0.300). Notably, in UHR acquisitions, increasing kernel sharpness did not lead to the expected SNR penalty: Hr76 demonstrated comparable or numerically higher SNR than Hr64 across dose levels (all p≥0.050). This counterintuitive pattern suggests a non-uniform, "bucket-like" interaction between reconstruction kernels and IR. Subjective analysis showed substantial inter-reader agreement (Fleiss κ=0.711) and consistent preference for sharper UHR reconstructions, even when acquired at lower radiation dose.
CONCLUSION: In midface PCCT, smaller detector pixels enable a measurable SNR advantage when matched for reconstruction sharpness. In addition, kernel-dependent interactions with IR may lead to non-monotonic noise behavior, allowing sharper reconstructions without the expected noise penalty.