2024 ARRS ANNUAL MEETING - ABSTRACTS

RETURN TO ABSTRACT LISTING


E4961. Cracking the Ionizing Radiation Paradox: Using Artificial Intelligence to Enhance the Clinical Utility of CT at a Reduced Radiation Cost
Authors
  1. Anthony Wu; Donald Bren School of Information and Computer Sciences, University of California, Irvine; School of Medicine
  2. Sungmee Park; School of Medicine
  3. Jinho Jung; School of Medicine
  4. Aidin Spina; School of Medicine
  5. Kenny Law; Department of Radiological Sciences, University of California, Irvine School of Medicine
  6. Parvaneh Hassani; Department of Radiological Sciences, University of California, Irvine School of Medicine
  7. Roozbeh Houshyar; Department of Radiological Sciences, University of California, Irvine School of Medicine
Objective:
The use of CT in medical care has steadily increased over the last 3 decades and has been linked with increased lifetime cancer risk. This risk was shown to be heightened for patients with repeat CTs, with up to an increase of 12%. Although it is important to note that CT ionizing radiation dosage (IoR) has decreased by a factor of 10 since 2009, lowering IoR in all imaging modalities while maintaining image quality remains a priority in radiology. Deep-learning image reconstruction algorithms (DLIRs) have been developed for this purpose. This systemic review examines the recent advances in DLIR’s efficacy in enhancing CT image quality at a reduced IoR.

Materials and Methods:
Database searches on PubMed, MEDLINE, and Google Scholar were performed. The last database search was obtained August 28, 2023. The literature search was limited to 2021–2023 to ensure relevance. The search strategy employed MeSH terms for imaging modalities, image reconstruction with machine learning, and lowering IoR. There were 54 records identified, of which 15 records (10 original studies) were of interest. Articles on DLIR not focused on IoR reduction were excluded based on title and abstract contents.

Results:
DLIR has been shown to maintain CT image quality at lower IoR across multiple organs. In the liver, studies have shown consistent hepatocellular carcinoma (HCC) conspicuity when using DLIR on either reduced CT or contrast dosage protocols(34% and 40%-50% IoR reductions, respectfully) compared to standard dosage and reconstruction(SDR) CTs. Another study showed a HCC protocol with both a 19.8% CT and 27% contrast dose reduction also preserved HCC conspicuity while increasing the contrast-to-noise ratio(CNR) and portal vein conspicuity.DLIR also shows promising results with cardiac CTs. When compared to SDR on cardiac CT angiography (CTA), one study showed that DLIR improved the signal-to-noise (SNR) and CNR by 50% on a cardiac CTA protocol using 40% less IoR. Another study showed DLIR usage in coronary CTA increased CNR by >60% across multiple coronary artery branches while reducing IoR by >50%.The utility of DLIR extends past solid organs. A study showed that in musculoskeletal(MSK) CT, DLIR was able to increase SNR by 47% on the paraspinal muscle while reducing IoR by 50%. In addition, the sharpness level of the vertebral bodies and psoas muscle were preserved in the same images. DLIR also successfully reduced CT-urogram (CTU) IoR, a protocol that generally requires three CT phases, by 70% compared to SDR CTUs.Certain populations may especially benefit from DLIR. Multiple studies show that DLIR significantly reduces noise in pediatric CTs using reduced IoRs. For bariatric patients who may require increased IoRs, DLIR was able to reduce CT IoR by 45% and contrast dose by 43% in coronary CTAs while reducing noise and increasing SNR and CNR when compared to SDR CTs.

Conclusion:
DLIR's efficacy in reducing CT IoR is shown across multiple organs. Populations such as pediatric and bariatric patients may benefit more from DLIR. Continued development is needed to further reduce IoR dosage and increase DLIR speed.