Diagnostic accuracy of AI-augmented renal ultrasound for degenerative kidney disorders: a systematic review and meta-analysis
TL;DR
Artificial intelligence-augmented renal ultrasound shows promising diagnostic performance, but heterogeneous populations, limited calibration, and predominantly internal validation constrain clinical generalizability.
Abstract
Abstract Introduction This systematic review and meta-analysis evaluated artificial intelligence (AI) and radiomics applied to renal ultrasound for the diagnosis, staging, and prognosis of degenerative kidney disorders. Methods PubMed/MEDLINE, Embase, Scopus, Web of Science, the Cochrane Library, and gray-literature sources were searched without date or language restrictions. Eligible studies applied AI or radiomics to renal ultrasound and reported diagnostic, staging, or prognostic outcomes. Risk of bias and methodological quality were assessed using the Prediction model Risk Of Bias Assessment Tool (PROBAST), Checklist for Artificial Intelligence in Medical Imaging (CLAIM 2024), and Radiomics Quality Score 2.0 (RQS 2.0). Random-effects models were used for diagnostic accuracy synthesis. Results Thirty-one studies were included. Machine-learning models achieved pooled sensitivity of 0.86 (95% confidence interval 0.82–0.90) and specificity of 0.83 (0.79–0.87); deep-learning models achieved sensitivity of 0.89 (0.84–0.93) and specificity of 0.85 (0.81–0.91). Heterogeneity was substantial and external validation was uncommon. Conclusions AI-augmented renal ultrasound shows promising diagnostic performance, but heterogeneous populations, limited calibration, and predominantly internal validation constrain clinical generalizability. Prospective multicenter external validation is required.
