Comparison of Seven Artificial Intelligence-Assisted Prediction Models for Renal Fibrosis in Chronic Kidney Disease Using Aggregate Index of Systemic Inflammation and Ultrasound Radiomics.

Objective: Renal fibrosis, key in progressive chronic kidney disease (CKD), requires invasive biopsy for diagnosis. This study aimed to develop an optimized artificial intelligence (AI)-assisted model for pre-biopsy screening.. Methods: A multicenter retrospective study included 758 renal fibrosis patients from two tertiary hospitals. 515 from The Second Hospital of Jingzhou were split 7:3 into training and internal validation sets; 243 formed an external test set. Severity (mild/moderate-to-severe) was classified via Banff score. Convolutional neural networks (CNN) extracted features from renal ultrasound images; peripheral blood counts were collected. After Least Absolute Shrinkage and Selection Operator (LASSO) variable screening, machine learning (ML) models were built, evaluated via receiver operating characteristic (ROC) curve's area under the curve (AUC) and decision curve analysis (DCA). Results: The model combining ultrasound radiomics and Aggregate Index of Systemic Inflammation (AISI) showed AUC 0.71-0.96 in training/internal validation, 0.89 in external test. Random Forest (RF) performed best (AUC 0.96 in training; 0.93/0.95 in internal/external validation). Conclusion: The RF-based model effectively evaluates renal fibrosis in CKD patients. Integrating AISI and ultrasound radiomics offers a novel strategy for dynamic assessment and biopsy guidance.