Kidney Week

Abstract: SA-PO0062

Deep Learning-Based Abdominal Aortic Calcification Assessment from Lateral Lumbar X-Rays Predicts Kidney, Cardiovascular, and Neurological Outcomes

Session Information

• Artificial Intelligence: Kidney Imaging, Pathology, and Precision Phenotyping
  October 24, 2026 | Location: Exhibit Hall A, Convention Center
  Abstract Time: 10:00 AM - 12:00 PM

Category: Artificial Intelligence, Digital Health, and Data Science

• 300 Artificial Intelligence, Digital Health, and Data Science

Authors

• Lu, Sz Ying, China Medical University Hospital, Taichung, Taichung City, Taiwan

Sz Ying Lu, MD

• Lin, Shuoyen, China Medical University Hospital, Taichung, Taichung City, Taiwan

Shuoyen Lin, MS

• Lin, Che-Chen, China Medical University Hospital, Taichung, Taichung City, Taiwan

Che-Chen Lin, MS

• Huang, Junxuan, China Medical University Hospital, Taichung, Taichung City, Taiwan

Junxuan Huang, MAS

• Chiang, Hsiu-Yin, China Medical University Hospital, Taichung, Taichung City, Taiwan

Hsiu-Yin Chiang, MS, PhD

• Kuo, Chin-Chi, China Medical University Hospital, Taichung, Taichung City, Taiwan

Chin-Chi Kuo, MD, PhD, MPH

Background

Abdominal aortic calcification (AAC) reflects systemic vascular calcification and has been associated with adverse cardiovascular and kidney outcomes. Although AAC is identified on routine lateral lumbar X-rays, manual assessment is time-consuming and not routinely performed. We evaluated whether deep learning (DL)-derived AAC predicts long-term kidney, cardiovascular, neurological outcomes, and mortality in a large real-world cohort.

Methods

We retrospectively included adults aged ≥50 years who underwent lateral lumbar X-rays at a tertiary medical center between 2008 and 2012. A DL pipeline using U-Net++ segmentation and ConvNeXt classification generated AAC-8 scores, with AAC positivity defined as AAC-8 ≥3 (Figure 1). Model performance was validated using 587 radiologist-annotated images. Patients with and without DL-derived AAC were matched by age, sex, and index year, resulting in 2,409 AAC-positive and 6,204 AAC-negative patients. Five-year risks of incident chronic kidney disease (CKD), end-stage kidney disease (ESKD), heart failure, coronary artery disease (CAD), stroke, dementia, and all-cause mortality were evaluated using multivariable Cox models.

Results

The DL model achieved an AUC of 0.937, sensitivity of 91.3%, specificity of 81.4%, and Cohen's κ of 0.734. AAC-positive patients had significantly increased risks of CKD (HR 1.28, 95% CI 1.09–1.51), heart failure (1.50, 1.23–1.83), CAD (1.46, 1.20–1.79), stroke (1.18, 1.02–1.36), dementia (1.24, 1.02–1.50), and all-cause mortality (1.36, 1.19–1.55; Figure 2).

Conclusion

DL-derived AAC assessment from routine lateral lumbar X-rays enables scalable opportunistic risk stratification and identifies individuals at increased risk of kidney, cardiovascular, neurological outcomes, and mortality in real-world clinical practice.

Figure 1. Representative examples of AAC-positive and AAC-negative images.

Figure 2. Hazard ratios (95% confidence intervals) for 5-year incident outcomes according to DL-derived AAC status.

Funding

• Government Support – Non-U.S.