SIR 2024
Imaging
Ahyeong Lee, BS
researcher
Seoul National University / ClariPi Inc., Republic of Korea
Financial relationships: Full list of relationships is listed on the CME information page.
Yongsu Bae, BS
Radiologic Techonologist
Seoul national university hospital, Republic of Korea
Disclosure information not submitted.
Saebeom Hur, MD, PhD (he/him/his)
Professor
Seoul National University Hospital, Republic of Korea
Financial relationships: Full list of relationships is listed on the CME information page.
jonghyo kim, PhD
professor
Seoul National University, Republic of Korea
Disclosure information not submitted.
This study collected low-dose (LD) C-arm CBCT datasets from a university hospital. The hospital conducted scans at a level one-fourth of the routine dose (RD) CBCT images for research purposes from December 1, 2019, to January 1, 2022. (RD DAP: s 5796.4 uGym² vs LD DAP: 1794.8 uGym²). The data was acquired from 90 patients who underwent transarterial chemoembolization (TACE) using Axiom Artis C-arm system (Siemens Healthcare, Forchheim, Germany). All CBCT images were reconstructed via filtered back projection (FBP) and were additionally reconstructed using ClariCT.AI (ClariPi Inc., Seoul, Korea), a DL-based denoising technique that was previously studied for use in MDCT images. Image quality comparison between denoised LD (DLD) and LD images was performed based on image noise and signal-to-noise ratio (SNR). The regions of interest (ROI) for quantitative evaluation abdominal aorta with diameter 1cm 2. After LD and DLD CBCT images were reconstructed MIP images, a radiologist assessed subjective noise, liver vessel visibility.
Results:
Image noises for DLD were 36.51 ±5.40 HU in liver parenchyma, 37.08 ±27.77 HU in vertebra erector spinae muscles and 53.20 ±11.33 HU in abdominal aorta, resulting in significantly lower noise level than those of LD which were 85.89 ±11.88 HU, 117.14 ±14.71 HU and 178.37 ±70.24 HU, respectively. SNR was 0.84 ±0.39, 0.65 ±0.40 and 0.59 ±0.16 for LD in liver parenchyma, vertebra erector spinae muscles and abdominal aorta, and was 1.97 ±0.89, 1.73 ±1.25 and 1.43 ±0.86 for DLD indicating superior level compared to LD. DLD MIP images showed reduced image noise and improved liver vessel visibility compared to LD MIP images.
Conclusion:
DL-based denoising can improve image quality in LD C-arm CT images during TACE. Implementing the DL-based denoising technique in C-arm CT can potentially lower patient radiation exposure during TACE procedures, a critical intervention for liver cancer treatment.