LAUSR

Background The development of artificial intelligence (AI) techniques has provided a novel strategy for improving the performance of renal ultrasound. To reflect the development of AI methods in renal ultrasound, we aimed to clarify and analyze the state of AI-aided ultrasound research in renal diseases. Methods PRISMA 2020 guidelines have been used to guide all processes and results. AI-aided renal ultrasound studies (for both image segmentation and disease diagnosis) published up to June 2022 were screened through the databases of PubMed and Web of Science. Accuracy/Dice similarity coefficient (DICE), the area under the curve (AUC), sensitivity/specificity, and other indications were applied as evaluation parameters. The PROBAST was used to assess the risk of bias in the studies screened. Results Of 364 articles, 38 studies were analyzed, and could be divided into AI-aided diagnosis or prediction related studies (28/38) and image segmentation related studies (10/38). The output of these 28 studies involved differential diagnosis of local lesions, disease grading of, automatic diagnosis, and diseases prediction. The median values of accuracy and AUC were 0.88 and 0.96, respectively. Overall, 86% of the AI-aided diagnosis or prediction models were classified as high risk. An unclear source of data, inadequate sample size, inappropriate analysis methods, and lack of rigorous external validation were found to be the most frequent and critical risk factors in AI-aided renal ultrasound studies. Conclusions AI is a potential technique in the ultrasound diagnosis of different types of renal diseases, but the reliability and availability need to be strengthened. The use of AI-aided ultrasound in chronic kidney disease and quantitative hydronephrosis diagnosis will be a promising possibility. The size and quality of sample data, rigorous external validation, and adherence to guidelines and standards should be considered in further studies.