Back to all papers

Physical image quality of normal- and super-resolution deep learning reconstruction in lung CT: a task-based phantom study.

November 19, 2026pubmed logopapers

Authors

Takada K,Urikura A,Shibata H,Urasaki S,Kuboyama T,Funasaka Y

Affiliations (7)

  • Department of Medical Technology, Ogaki Municipal Hospital, 4-86 Minaminokawa-cho, Ogaki City, Gifu, Japan. [email protected].
  • Department of Radiological Sciences, Ibaraki Prefectural University of Health Sciences, Ibaraki, Japan. [email protected].
  • Department of Radiological Sciences, Graduate School of Ibaraki Prefectural, University of Health Sciences, Ibaraki, Japan.
  • Department of Diagnostic Radiology, National Cancer Center Hospital, Tokyo, Japan.
  • Department of Diagnostic Radiology, Fujita Health University School of Medicine, Aichi, Japan.
  • Department of Radiological Technology, Toyota Kosei Hospital, Toyota, Aichi, Japan.
  • Department of Medical Technology, Ogaki Municipal Hospital, 4-86 Minaminokawa-cho, Ogaki City, Gifu, Japan.

Abstract

To quantitatively compare the physical image quality characteristics of normal-resolution deep learning reconstruction (NR-DLR) and super-resolution deep learning reconstruction (SR-DLR) in lung CT, with particular emphasis on task-based detectability. A 180-mm TOS-SS phantom was scanned at four dose levels (CTDI<sub>vol</sub>: 11.2, 5.6, 2.6 and 1.0 mGy). Images were reconstructed using NR-DLR (AiCE) and SR-DLR (PIQE) at three noise reduction levels (Levels 1-3), with 512- and 1024-matrix settings at display fields-of-view (D-FOVs) of 320 and 180 mm. Spatial resolution was assessed using the task transfer function (TTF), noise characteristics were evaluated using the noise power spectrum (NPS), and task performance was quantified using the non-prewhitening detectability index (d') for a 10-mm, 130-HU disk target. An additional analysis was performed using a 3 mm, 130 HU target to evaluate task dependence. AiCE demonstrated higher f₅₀ values and characteristic frequency enhancement (overshoot), whereas PIQE exhibited a monotonic TTF profile and lower high-frequency noise components. Despite lower f₅₀ values, PIQE yielded higher d' values in most reconstruction settings for the 10 mm task. For the 3 mm task, the difference in d' between PIQE and AiCE was reduced, and their relative relationship was reversed in some reconstruction settings. NR-DLR and SR-DLR demonstrated distinct resolution-noise characteristics in lung CT. Despite lower f₅₀ values, SR-DLR yielded higher task-based detectability in most reconstruction settings for the 10 mm task; however, the relative detectability performance of NR-DLR and SR-DLR was task dependent.

Topics

Journal Article

Ready to Sharpen Your Edge?

Subscribe to join 11k+ peers who rely on RadAISlice. Get the essential weekly briefing that empowers you to navigate the future of radiology.

We respect your privacy. Unsubscribe at any time.