Back to all papers

Leveraging CT-derived chronic imaging signatures for acute kidney injury evaluation.

June 13, 2026pubmed logopapers

Authors

Kim K,Choi YH,Lee HG,Kim B,Paek JH,Hong JU,Lee RW

Affiliations (8)

  • Division of Nephrology, Department of Internal Medicine, CHA Bundang Medical Center, CHA University School of Medicine, Seongnam, Republic of Korea.
  • Department of Biomedical Informatics and AI Center, Hallym University Chuncheon Sacred Heart Hospital, Chuncheon, Republic of Korea.
  • Department of Electrical and Computer Engineering, Inha University, Incheon, Republic of Korea.
  • College of Medicine, Inha University, Incheon, Republic of Korea.
  • Department of Radiology, Keimyung University School of Medicine, Daegu, Republic of Korea.
  • Department of Internal Medicine, Keimyung University School of Medicine, Daegu, Republic of Korea.
  • Department of Radiology, Inha University Hospital, Inha University College of Medicine, 27 Inhangro, Jung-Gu, Incheon, 22332, Republic of Korea.
  • Department of Radiology, Inha University Hospital, Inha University College of Medicine, 27 Inhangro, Jung-Gu, Incheon, 22332, Republic of Korea. [email protected].

Abstract

Accurate estimation of baseline serum creatinine (SCr) remains an unmet clinical need in acute kidney injury (AKI) evaluation, as premorbid SCr values are often unavailable. This study aimed to develop and validate a baseline SCr prediction model using CT-derived imaging features to improve AKI diagnosis and staging. The study included patients with available baseline SCr who underwent abdominal CT at two tertiary hospitals. Kidney segmentation was performed using a fine-tuned Swin UNETR model, from which imaging features were extracted. We evaluated machine learning models incorporating imaging features for baseline SCr estimation and compared their estimates with those derived from the back-calculation method. Most selected imaging features were associated with the presence of preexisting chronic kidney disease. Across internal and external test sets, the tabular foundation models achieved the best performance (MAE, 0.154-0.158/0.168-0.174; RMSE, 0.253-0.261/0.220-0.231). AKI staging based on predicted baseline SCr showed substantially higher agreement with the ground truth than back-calculation (Cohen's κ, 0.69/0.60 vs. 0.31/0.36), while markedly reducing over-staging (8.7%/8.5% vs. 45.2%/37.6%). Our externally validated, image-driven prediction model enabled accurate baseline SCr estimation and improved AKI classification.

Topics

Journal Article

Ready to Sharpen Your Edge?

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

We respect your privacy. Unsubscribe at any time.