This article provides a formal correction to the study "A noninvasive model for chronic kidney disease screening and common pathological type identification from retinal images," originally published in Nature Communications (July 2025). The correction addresses graphical errors in the representation of model performance metrics without altering the study's core findings or scientific integrity.
Background: The Retinal-Renal Axis
The original research addressed a critical gap in nephrology: the need for noninvasive and accessible screening methods for chronic kidney disease (CKD). Traditionally, diagnosing CKD and identifying specific pathological types requires invasive procedures, including blood tests, urine analysis, and kidney biopsies.
By leveraging the "retinal-renal axis"—the biological and microvascular link between the eye and the kidney—the authors developed a deep-learning model designed to identify CKD and its pathological variations using simple retinal fundus images. This approach aims to provide a low-cost, noninvasive alternative to traditional diagnostic pathways.
The Nature of the Correction
The primary purpose of this amendment is to rectify a clerical error in the visual representation of the model's performance. In the version of the article initially published, a figure preparation error led to the inadvertent duplication and misplacement of several Receiver Operating Characteristic (ROC) curves and Area Under the Curve (AUC) values within Figure 3 (specifically panels a–e).
ROC curves and AUC values are standard metrics used to evaluate the diagnostic accuracy of a model. While these visual elements were incorrectly placed in the initial layout, the authors have clarified that the underlying data and the statistical analysis used to generate these metrics were correct from the outset.
Scientific Integrity and Conclusions
It is important to emphasize that these graphical errors do not affect the original data, the analysis, or the scientific conclusions of the study. The fundamental findings regarding the efficacy of the AI model in screening for CKD and identifying pathological types remain sound.
The publication of this correction serves as a formal update to the scientific record. The figure has been updated in both the HTML and PDF versions of the article. For full transparency, the original erroneous figure remains available as Supplementary Information alongside the amendment, allowing for a direct comparison of the rectified data visualization.
This correction ensures that the published performance metrics accurately reflect the model's capabilities as described in the research, maintaining the transparency and rigor expected in clinical AI research.
Summary updated