CT-based renal and body-composition radiomics model to improve the detection ability of diabetic kidney disease in patients with type 2 diabetes mellitus

Figure 1. (A) Flowchart of the study. (B) The workflow of this study. Images were obtained and semi-automatically segmented on 3D slicer. We then utilized the Shukun platform to extract and preselect radiomics features. The resultant features and clinical variables were separately screened to establish radiomics, clinical models and a combined model. Nomograms, ROC curves, calibration curves, and decision curve analyses were performed for model assessment. ROC, receiver operating characteristic; T2DM, type 2 diabetes mellitus; CT, computed tomography; DKD, diabetic kidney disease.

Figure 2. A schematic diagram showing modules of interest. (A) renal sinus adipose areas at the level of renal hilum; (B) kidney areas at the level of renal hilum; (C) perirenal adipose areas at the level of renal hilum; (D) renal sinus adipose volumes; (E) kidney volumes; (F) perirenal adipose volumes; (G) visceral fat area at the 3rd lumbar vertebra level; (H) subcutaneous fat area at the same level; (I) skeletal muscle area at the same level; (J) reconstruction view of visceral fat volume between diaphragm and sacrum; (K) subcutaneous fat volume; (L) skeletal muscle volume.

Figure 3. Performance of the radiomics model. (A) Coefficient paths across log(λ) values in the LASSO regression. (B) 10-fold cross-validation of the LASSO regression. (C) Nomogram for selected radiomics features of DKD vs. no-DKD in patients with T2DM. C, RSATArea-auto:original_shape_Elongation; D, RSATArea-auto:wavelet-HHL_glszm_SmallAreaLowGrayLevelEmphasis; G, RenalArea-auto:lbp-3D-k_glszm_SmallAreaEmphasis; AG, PRATVolume-auto:wavelet-HHH_glszm_SmallAreaHighGrayLevelEmphasis; AH, PRATVolume-auto:wavelet-HLL_firstorder_Maximum. (D) ROC curves of the radiomics model in the training, internal and external validation cohorts. The red line represents the training cohort curve. The blue line represents the internal validation cohort curve, whereas the green line represents the external validation cohort curve. (E) Calibration curves of the nomogram for the external validation cohort. (F) Analysis of clinical net benefit of the radiomics model across different decision thresholds to guide interventions in the external validation cohort. LASSO, least absolute shrinkage and selection operator; RSAT, renal sinus adipose tissue; PRAT, perirenal adipose tissue; ROC, receiver operating…

Figure 4. The performance of the clinical model. (A) Nomogram for clinical features of DKD vs. no-DKD in patients with T2DM. (B) ROC curves of the clinical model in the training, internal, and external validation cohorts. The red line represents the training cohort curve. The blue line represents the internal validation cohort curve, whereas the green line represents the external validation cohort curve. (C) Calibration curves of the nomogram for the external validation cohort. (D) Analysis of clinical net benefit of the clinical model across different decision thresholds to guide interventions in the external validation cohort. ROC, receiver operating characteristic; T2DM, type 2 diabetes mellitus; DKD, diabetic kidney disease.

Figure 5. The performance of the combined model. (A) Nomogram for integrated variables of DKD vs. no-DKD in patients with T2DM. C, RSATArea-auto:original_shape_Elongation; D, RSATArea-auto:wavelet-HHL_glszm_SmallAreaLowGrayLevelEmphasis; G, RenalArea-auto:lbp-3D-k_glszm_SmallAreaEmphasis; AG, PRATVolume-auto:wavelet-HHH_glszm_SmallAreaHighGrayLevelEmphasis; AH, PRATVolume-auto:wavelet-HLL_firstorder_Maximum. (B) ROC curves of the combined model in the training, internal, and external validation cohorts. The red line represents the training cohort curve. The blue line represents the internal validation cohort curve, whereas the green line represents the external validation cohort curve. (C) Calibration curves of the nomogram for external validation cohort. (D) Analysis of the clinical net benefit of the clinical model across different decision thresholds in the external validation cohort. RSAT, renal sinus adipose tissue; PRAT, perirenal adipose tissue; ROC, receiver operating characteristic; DKD, diabetic kidney disease.

Figure 6. Comparison of features derived from two- and three-dimensional regions on the training (A), internal validation (B), and external validation (C) cohorts. Comparative analysis by DeLong’s test is displayed in (D). 2D, two-dimension, 3D, three-dimension. .