Kidney Week

Abstract: SA-PO954

Radiomic Features of Kidney Magnetic Resonance Imaging (MRI) to Characterize CKD and Progressive CKD

Session Information

• CKD: Observational Research and Patient-Oriented Interventions
  November 05, 2022 | Location: Exhibit Hall, Orange County Convention Center‚ West Building
  Abstract Time: 10:00 AM - 12:00 PM

Category: CKD (Non-Dialysis)

• 2202 CKD (Non-Dialysis): Clinical‚ Outcomes‚ and Trials

Authors

• Leidner, Alexander S., Northwestern University Feinberg School of Medicine, Chicago, Illinois, United States

Alexander S. Leidner,

• Li, Luping, NorthShore University HealthSystem, Evanston, Illinois, United States

Luping Li,

• Wilt, Emily, NorthShore University HealthSystem, Evanston, Illinois, United States

Emily Wilt,

• Sprague, Stuart Michael, NorthShore University HealthSystem, Evanston, Illinois, United States

Stuart Michael Sprague,

• Rusinek, Henry, NYU Langone Health, New York, New York, United States

Henry Rusinek,

• Srivastava, Anand, Northwestern University Feinberg School of Medicine, Chicago, Illinois, United States

Anand Srivastava,

• Prasad, Pottumarthi V., NorthShore University HealthSystem, Evanston, Illinois, United States

Pottumarthi V. Prasad,

Background

CKD represents a wide-spectrum of conditions with varying etiologies, manifestations and rates of progression. Application of radiomics, objective numerical representations of image texture, on MRI has potential to non-invasively identify disease features through evaluation of spatial heterogeneity. However, there is limited data on the application of kidney MRI-based radiomics in individuals with CKD.

Methods

We generated radiomic features from arterial spin labeling (ASL)-derived cortical blood flow (BF), blood oxygenation level dependent (BOLD)-derived relaxation rate (R2*), and diffusion-weighted (DW)-derived apparent diffusion coefficient (ADC) maps on 9 healthy and 24 individuals with diabetic CKD (eGFR <60 ml/min/1.73m²). FireVoxel was used to calculate 54 radiomic features of the kidney cortex. Spearman correlations were calculated between features. Logistic regression models of radiomics and clinical variables were used to classify (1) CKD and (2) rapidly progressive CKD (eGFR decline >3 ml/min/1.73m²/year). Forward stepwise regression was based on area under the receiver operating characteristic curve (AUC) changes.

Results

The best model in predicting CKD used the ASL-derived mean BF value had an AUC of 1.0. The best model to predict rapidly progressive CKD included ASL-derived BF and BOLD-derived R2* parameters with an AUC of 1.0 (Figure 1a). Model parameters were significantly correlated with the presence of CKD, eGFR, eGFR slope, and urine protein excretion (Figure 1b).

Conclusion

These preliminary results show the potential of MRI radiomics as a non-invasive diagnostic tool in phenotyping CKD. These results require validation in future larger studies with heterogeneous CKD etiologies.

Funding

• NIDDK Support