Abstract: TH-PO026
Development of a Multimodal "Kidney Age" Prediction Based on Automatic Segmentation of Magnetic Resonance Imaging in Individuals with Normal Kidney Function: A Preliminary Report
Session Information
- Augmented Intelligence for Prediction and Image Analysis
October 24, 2024 | Location: Exhibit Hall, Convention Center
Abstract Time: 10:00 AM - 12:00 PM
Category: Augmented Intelligence, Digital Health, and Data Science
- 300 Augmented Intelligence, Digital Health, and Data Science
Authors
- Hou, Zuoxian, Peking Union Medical College Hospital, Beijing, China
- Ma, Yixin, Peking Union Medical College Hospital, Beijing, China
- Xu, Lubin, Peking Union Medical College Hospital, Beijing, China
- Zhang, Gu-Mu-Yang, Peking Union Medical College Hospital, Beijing, China
- Xia, Peng, Peking Union Medical College Hospital, Beijing, China
- Chen, Limeng, Peking Union Medical College Hospital, Beijing, China
Background
Traditional kidney health assessments rely on clinical indicators and demographics. We introduced KAGE-NET, an artificial intelligence model, to predict kidney age (K-AGE) by combining auto-segmentation and radiomics features from renal MR images to understand kidney health better.
Methods
MR images (T1-weighted and T1-mapping) from UK Biobank, comprising 5693 participants, were utilized in four groups (healthy control (Normal), hypertension (HTN), diabetes (DM) and chronic kidney disease (CKD)). All data underwent auto-segmentation and image-derived phenotype extraction. Control group data was divided into an 8:2 ratio for KAGE-NET training and testing. In the disease groups, the kidney age gap (KAG) metric (K-AGE minus biological age) described the acceleration of kidney aging. KAG greater than 0 indicates accelerated kidney aging; vice versa for KAG less than 0.
Results
A total of 3007 female and 2682 male subjects were included, with a mean age of 53.7±7.8 years old and sCr of 72.28±13.92 μmol/L. Ground truth (Fig.1a) and auto-segmentation (Fig.1b) show high correlation. KAG is around 0 in control group but significantly different with disease groups (CKD: 1.16; DM: -0.58; HTN: -0.25; p<0.01; Fig.1c). As health conditions worsen, KAG widens significantly, especially in more complex situations (Fig.1d). Essentially, when DM co-exists without/with CKD, a heightened KAG comparison is noticeably evident (-0.75 vs. 1.52, Fig.1e). A similar outcome is detectable in the KAG comparison between HTN without/with CKD (-0.35 vs. 1.41, Fig.1f).
Conclusion
We introduced KAGE-NET based on MRI multimodal data, presenting a promising comprehensive kidney health assessment tool.
(a-b) An auto-segmentation example. (c-f) KAG between groups.
Funding
- Government Support – Non-U.S.