Integrated machine learning and deep learning for predicting diabetic nephropathy model construction, validation, and...

Area of Science:

• Nephrology
• Artificial Intelligence
• Medical Informatics

Background:

• Diabetic Nephropathy (DN) is a significant complication of diabetes, necessitating early and accurate diagnostic tools.
• Current diagnostic methods can be invasive or lack sensitivity for early-stage detection.

Purpose of the Study:

• To develop and validate a machine learning-based risk prediction model for assisted diagnosis of Diabetic Nephropathy (DN).
• To evaluate the model's predictive performance and generalizability through internal and external validation.

Main Methods:

• Data preprocessing and feature selection using algorithms like LASSO, RFE, and MRMR.
• Construction and evaluation of ten machine learning models, with Random Forest identified as optimal.
• Internal and external validation using metrics including ROC, PR, Accuracy, MCC, and Kappa.

Main Results:

• Fifteen key variables were identified for DN prediction.
• The Random Forest model exhibited superior predictive performance with an ROC of 0.912 in the test set.
• External validation cohorts showed strong generalization with ROC values of 0.828 and 0.863.

Conclusions:

• The developed machine learning model demonstrates significant predictive value for Diabetic Nephropathy.
• This model is poised to assist in the early diagnosis and clinical screening of DN.
• An online platform based on the model facilitates accessibility for clinical application.